In this episode of Class Disrupted, Netflix founder Reed Hastings joins Michael Horn and Diane Tavenner to discuss his decades-long journey through various chapters of education reform and how it鈥檚 shaped his view around artificial intelligence shaping the space. Reflecting on the slow progress and setbacks of past education initiatives, the episode dives into the potential of and urgency for harnessing AI for accelerated, mastery-based learning and global impact. Reed shared what he believes reinventing traditional classrooms means for edtech entrepreneurs.

Listen to the episode below. A full transcript follows.

Michael Horn: Remember how much fun it was to have Reed Hastings join us on Class Disrupted in the beginning of the season to dish on AI and education? So much fun that the ASU + GSV Summit said, why don鈥檛 you invite him back and let鈥檚 do it again. So that鈥檚 what we did. Six months later, Diane Tavenner and I welcomed Reed Hastings live on stage at the ASU + GSV Summit to talk AI, his different chapters in education, lessons he鈥檚 learned, and where he thinks the puck is going. Enjoy all of that on this episode of Class Disrupted live from the ASU + GSV Summit and sponsored by the Learner Studio.

Michael Horn: Welcome everyone to the Class Disrupted podcast. This is our seventh season doing it. Normally we are disembodied voices on a screen talking with each other and a guest, but tonight we鈥檝e got a live audience and we want to thank the ASU + GSV Summit and all of the amazing staff that has put this on. Huge thanks for all of them. Please. And you are all here to make a lot of noise and make this fun, right? Diane?

Diane Tavenner: Indeed. That鈥檚 what we want, is a spirited conversation.

Michael Horn: So who do we鈥檝e got on tap?

Diane Tavenner: Well, tonight, Michael, we have an incredible guest, someone we鈥檝e talked to before, but it is time to do it again. We have Reed Hastings with us. And most people know Reed from Netflix. A lot of people know that he spent time in education. What you might not know is that for the last year, Reed鈥檚 been on the board of Anthropic. He鈥檚 done a deep, deep dive in AI, the recent version, and 40 years ago earned a master鈥檚 in AI.

Michael Horn: So Reed, you鈥檝e also had a long set of experiences with education over the years. You were a Peace Corps member, teaching maths 40 some-odd years ago, I think. And 20 some-odd years ago, you were the chair of the California State Board of Education. A testing period, No Child Left Behind. You had this guy named Roy Romer in Los Angeles as the superintendent for some seven years. What did you learn from that period of time working in education?

Leadership changes in education systems

Reed Hastings: Well, that was a time of great hope. We had No Child Left Behind, Reading First, high school exit exams. We had an accountability system and I was really an administrator of that on the State Board of Education. And we worked hard on all the technical details and there was some real progress. And as you mentioned, Roy Romer was very successful as superintendent. He made it seven years in LA Unified as superintendent, set a record for that and put in a lot of great programs that really raised scores and achievement, learning. And the tragic thing was the next five years after that, I watched it all get dismantled, independent of its results. It sort of, you know, politically wasn鈥檛 in favor.

New administrations elected, they are like, get rid of the old guy stuff and let鈥檚 put in different stuff. And so that was true at the school district level and that was true at the state level. And it really woke me up to the hero syndrome we have. And whether it鈥檚 Tom Pazant鈥檚 great work in Boston now getting dismantled or Houston, Mike Miles is the hero today. And you know, watch what will happen in five or 10 years from now, because that鈥檚 where Rod Page was so great 40 years ago. Of course there was Joel Klein in New York that so many people worked hard on. So we see this cycle of rise and fall. And I have to say, for all the work that I did and all that state board, there鈥檚 very little to show for it.

Diane Tavenner: So in another chapter where we met, was in the charter world, and you鈥檝e been on the board of KIPP national for 20 years now. You have supported countless of us who have been in the work throughout that time, you know, the City Fund. And this is a long-term strategy for you. What are you learning from charters?

Reed Hastings: Well, I would say charters haven鈥檛 failed, but they haven鈥檛 succeeded at driving up NAEP scores in the high charter states, say like Arizona, Texas, Florida. And of course unions have fought us to a draw in deep blue states and then in red states we鈥檙e able to grow and we鈥檙e investing in, again, Florida, Texas, Arizona, Georgia, so lots of states. But even, you know, after 20 years, we have a good success at the city level. So at the city level, it鈥檚 actually the only thing that鈥檚 driven citywide improvement for all kids is high charter share. So if you look, PPI did the graph, the scatter plot showing that cities with low charter share, Portland, Seattle, have had no improvement in closing the gap of achievement between poor kids and all kids over the last 25 years. And then you start walking up the cities that have 10% charter share, more improvement, 20, 30, 40, 50, like Newark, Camden. And then you get to New Orleans, which has the highest gap closure in the nation over the last 25 years. And of course that鈥檚 100% charter.

So charter is still promising, but like grindingly hard and slow. Think trench warfare, but it hasn鈥檛 been reversed. OK, so a lot of positivity and I continue to be a huge donor in that space and continue to believe in it on maybe a half dozen boards of charter networks.

Michael Horn: So the third chapter that you then went in on an education is when I met you, 2010, the very first ASU GSV summit, you were there and you were getting involved in education technology, EdTech and DreamBox Learning, of course. And there鈥檚 been a whole wave, sort of cresting, if you will, with EdTech. What鈥檚 your take on that chapter?

Reed Hastings: Yeah, well, Rocketship was using DreamBox Learning, and I knew it through there, and I thought, OK, here鈥檚 a great opportunity to take this amazing software. And obviously computers transform everything. And so if we could just get some investment in DreamBox and get it bigger, it would surely transform both district schools and charters. And again, grindingly slow. Turns out that selling to school districts is really hard. The only thing harder is selling to charters because they鈥檙e small, so the money鈥檚 on the district side, but grindingly so. And then, you know, DreamBox was one of the early adaptive learning, you know, let kids go at their own pay systems.

But school districts kept telling us to turn that off, please, because they wanted to catch kids up to grade level, but they definitely didn鈥檛 want to get kids ahead because if the kid gets ahead, then they鈥檙e disruptive and bored in the class. So catching kids up to make the machine work better, very much valued. Letting kids get ahead, which sort of threw sand in the machine, not valued. And so it was an early lesson in sort of the depth and strength of the grammar of schooling that we have.

Diane Tavenner: So if I sum up these three chapters, state policy, district work, pop of success gets wiped away. Charters making progress haven鈥檛 failed, grindingly slow ed tech, no real discernible change yet. I know you鈥檙e not trying to depress us. I know you鈥檙e trying to help us know that you鈥檙e learning and still in the game, which we know you are. So let鈥檚 get back to AI. When鈥檚 it going to cure cancer? When is it going to figure out fusion so energy is free? When is it going to autonomously drive us all over the place so we don鈥檛 have to deal with parking lots anymore? When is it going to make our lives better?

Rapid AI advancements predictions

Reed Hastings: By the end of the summer? Predicting AI is tricky because it鈥檚 growing so fast in quality. You know, it was three years ago when ChatGPT came out and it could barely do third grade math. And now all of the major AI systems are very impressive and they鈥檒l continue to improve. And what鈥檚 happening is we鈥檙e on one of these curves where it鈥檚, let鈥檚 call it doubling every year in quality. So it will be twice as good as it is today a year from now, and then twice as good, and then twice as good and then twice as good. So whatever challenge you think AI is not up to, just wait a year. OK? And so that鈥檚 the amazing thing. And there鈥檚 no guarantee that the exponential will continue forever, but it has been the last several years and you know, it鈥檚 getting very, very impressive at many scenarios like the ones you talked about and many others.

So the amount of change that we鈥檙e going to see in our society, mostly positive, but there鈥檒l be some negative, from AI getting better and better is hard to grasp because of this doubling, doubling, doubling. You know, just when you think we鈥檝e got it like situated like how鈥檚 it going to work with society? Then it gets even better again. And so we鈥檙e in for the ride of our lives, both on the positive side. So curing cancer, energy, you know, abundance, these kinds of things and on the stress side of everything is different than it was when we grew up.

Michael Horn: Well, that鈥檚 the question I want to ask you because not only is there this anxiety and stress, as you know, people are also worried, will people get hurt as it gets better? And you know, you can imagine a myriad of ways that could play out. What鈥檚 your take on how do we prevent people from getting hurt?

Reed Hastings: Yeah, and I mean, again, that鈥檚 happened with some tragic cases of, you know, teens and suicide already. And look at the societal level, we make certain choices, sometimes explicitly, sometimes implicitly. And we tend to accept the choices that are already made for us and be scared, scared about new ones. But for example, you know, we lose 40,000 people a year to car accidents in the U.S. and about half a million globally. And if we just ban cars, you know, we wouldn鈥檛 have those deaths. OK, but we鈥檙e not willing to pay the price. So implicitly we鈥檙e making a trade off of 40,000 U.S. deaths a year. So I look at it and say, you know, is it as powerful as a car? And if it is, then I鈥檓 like, I know where society is in making those trade offs.

So I don鈥檛 want to pay that price. I don鈥檛 want to see 400,000 or 40,000 a year deaths. But I think when we get all excited about four deaths, we鈥檙e sort of losing perspective about the size of the prize and the other trade offs that we have and continue to make in society. So, you know, AI, I think will reduce deaths, and in particular with self-driving cars, that should be able to eliminate 90% of those 40,000 US deaths through self driving if we can get that adopted. OK, but then you see the story of the one Tesla death that happens. And again, that death鈥檚 tragic. I鈥檓 not trying to take away from it, of course, but in comparison to all the lives that self driving is already saving, it鈥檚 quite small.

Diane Tavenner: So let鈥檚 take that into education now, because one of the things that I love about you is that you keep learning and you stay in the work when a lot of people leave, and I know that there is a fourth chapter that is going to be written in your work and it鈥檚 going to involve AI. And so what does education look like in the age of AI? What does school look like in the age of AI? What does learning look like in the age of AI?

Improving education over the years

Reed Hastings: Yeah, but in my first 25 years, I鈥檝e spent the time trying to do the better classroom, whether that鈥檚 from the state board level and testing and assessment, how do we make schools and classrooms better, whether that鈥檚 using ed tech like DreamBox Learning to make the classroom better. Charter schools, which have had some progress in making the classroom better. But it reminds me of the story about steam powered factories in the 1800s. So in the 1800s, all of our factories had a big steam plant that burned coal and rotated an engine. And then throughout the factory we had a rotating rod which carried power through the plant. And then we had belts and pulleys and wheels that then spun the individual looms or other machines. And these were highly developed, mechanized, and lots of belts and pulleys throughout the factory. You know, lot of productivity.

Then electricity comes and we replace the big steam engine with a big electric engine. And that saves some money. But real productivity of the factories didn鈥檛 change. And this puzzled economists for a long time. And then people started saying, hey, the power distribution system, all those pulleys and rods spinning, that鈥檚 the problem. And if we get rid of that and then go to individualized electric motors, so each loom has its own motor, then it can be designed sideways because the power is not all in one direction. Then it鈥檚 variable speed. You can turn off some motors and turn on other ones and all of these subtle effects.

Then we had a huge increase in factory productivity from basically using electricity the way it should be used in lots of small, relevant motors, rather than replace the one big motor. And I remember hearing that story and thinking, oh my gosh, that鈥檚 what鈥檚 happening in education. We鈥檙e putting tech into the classroom and the classroom, the sage on a stage, is the power distribution system. The sage on a stage is holding back technology from its natural effects and its ability to teach children directly. And we have to be brave enough to try to do school without sage on a stage at all. OK? To have all of school be learning individually, your daily lesson plan from the system executing.

Experimenting with individualized tutoring

Reed Hastings: We want to maintain the social development so the person in the classroom really becomes a social worker. They鈥檙e specializing in learning and emotional maturity and doing valor-type circles and these kinds of things. But the quote “education learning” stuff all becomes individualized where it鈥檚 mastery based learning. And the question is, how much more would kids learn? So one experimental way to get at this is to think about Bloom 40 years ago, and Bloom said two sigma improvement from individual tutoring, but it hasn鈥檛 been revalidated in a large scale way in a while. And, and so one of the projects we鈥檙e doing is funding that and you know, take 50 random kids, median kids in a median school, and give them a full year of the whole school day individual tutoring and try to figure out, OK, how much more do they learn? And so Ben Rosen, who鈥檚 here at the conference, runs Recess.gg, he鈥檚 running this project and recruiting tutors. And so let鈥檚 see, for second graders in the ideal condition, how much can they learn? What is the rate of learning of typical human 7 year olds? And I think we鈥檙e going to see it鈥檚 a whole lot faster than one grade level in one year, when again, completely individualized tutor, they can do everything moral and legal. They want to help the kid learn more in that year. All kinds of motivational things, all kinds of different teaching techniques.

But again, it鈥檚 one on one, dedicated. And you might say, well look, you know, that鈥檚 so expensive, $100,000 per kid per year. It鈥檚 ridiculous. And I would say that鈥檚 what it is now. But with AI, it gives all the AI developers a target of what they鈥檙e trying to do and how much more learning. And what we want the world to understand is, no, there really is twice as much learning that could be happening per day, per hour than today, because I suspect that we鈥檒l find that it is twice as much, which roughly means by the time you get to eighth grade, you know as much as a typical high schooler today or by the time you get to 11th grade, you know, as much as the typical college student today. OK? Because of the time compression and the learning and the stimulation.

And that would lead to, you know, not just lifting the bottom, which of course it does, but just a tremendous revolution in the possibilities of the human brain. And there鈥檚 a positive example of this. So about 25 years ago, Deep Blue beat Garry Kasparov in chess. And from then on, AI chess has been better than human chess. And so you might think, well, everyone stopped playing chess and it鈥檚 kind of gotten irrelevant. But in fact, chess has grown. And now the typical 10 year old on Chess.com is scoring way higher than the 10 year olds of 20 years ago on a stable, vertically scored system. And what鈥檚 happening is the 10 year olds are getting tutored by AI and the 12 year olds and 14 year olds.

And so we鈥檙e seeing this rise in chess talent because they鈥檙e individually tutored by AI. And so that鈥檚 true for chess today and could be true for biology and history tomorrow.

Diane Tavenner: I know Michael has a lot of questions, but before we just move, hold, hold. Because I don鈥檛 want this to get lost. And I think people often get confused when we talk about the power of individual tutoring. And they think kids are going to be learning by themselves. And that is not what you鈥檙e saying here. I know that鈥檚 not what you鈥檙e saying.

Reed Hastings: A dark room, nothing there, locked in. We can reuse containers. No, you want all the social development that we have today. So it鈥檚 real.

Diane Tavenner: Because those chess kids are playing chess with other kids.

Reed Hastings: That鈥檚 right. And if you just take the chess, if you just take the school day and say the time that鈥檚 direct instruction, sage on the stage now becomes individualized tutoring. And all the play time and all the time that鈥檚 do a project together stays as that. And in fact you can be. The teachers can then focus on that aspect of the day. And again, social, emotional learning, we all know is important. But imagine if the teacher鈥檚 an expert in it and focuses on that because understanding and doing well on the stuff that鈥檚 tested is done by the software.

Diane Tavenner: And by the way, sage on the stage is a very lonely experience anyway, so let鈥檚 not pretend.

Michael Horn: Speaking from experience. Well, I was gonna say you鈥檙e gonna finally disrupt class, which I鈥檓 thrilled by, but yes. But I鈥檓 curious because I talk to a lot of ed tech entrepreneurs at this conference and elsewhere. What鈥檚 your advice to them? Because they do a lot of times what DreamBox did, right, which is sell to the existing system, the districts, the schools, the sage on the stage. What鈥檚 your advice to them?

Reed Hastings: Yeah, it鈥檚 a great point. The short term is if you want to make money selling to school districts, make teachers鈥� lives easier. OK, don鈥檛 worry about learning too much. But if you make teachers鈥� lives easier, you鈥檒l sell well. If you want to change the world, focus on the homeschoolers. Focus on people who are able to go at their own pace and build systems that are individualized. And as the benefits of that are more and more clear, not meaning 5%, but meaning twice as much learning, school districts will move towards that.

Self-learning education technology

Reed Hastings: And so if you build that now, you鈥檙e skating to where the puck is going, which is this individualized education. And so think of it as trying to do the pure play where you don鈥檛 need a teacher. OK? It is the self driving car where most of the market is like the map in the car to help the human. OK? That鈥檚 where most of our ed tech is. And instead we need to build the self-driving car in terms of innovation, which is the self learning, self teaching. And again, the AI is getting better and better at the emotional motivation.

So when you, you know, the vast majority of people seeking therapy today are getting therapy from chat, not from waiting a week and going and seeing someone at 80 bucks an hour. It vastly expanded the market. And you can say, well, it鈥檚 uncertified and that鈥檚 all true, but it is satisfying to people and it鈥檚 not perfect in any way. It is getting better and better rapidly back to that doubling. OK? And so the understanding, the emotional nuance of humans is something that actually the software is, is quite good at and getting better.

Diane Tavenner: And we could talk for days and days about how this leads to agency and self direction and entrepreneurial spirit and when they鈥檙e getting what they need.

Reed Hastings: Yeah, once you learn how to learn from software and from the interaction, the world鈥檚 your oyster because then you go off and you want to do physics or you want to do history again, a lot of it is there.

Diane Tavenner: So before. Yeah, let鈥檚 take it to the world. So what does this mean to the world you are working globally? CJ is here in the audience with us. Tell us about your work in Africa.

Sharing AI education globally

Reed Hastings: Yeah, it鈥檚 one of the most exciting secondary effects of this AI revolution is it鈥檚 very shareable when we figure out good teaching practices like Success Academy or KIPP, it鈥檚 very hard to export that to a Brazilian or African context. But when you figure out tech, it鈥檚 very easy to share. So, you know, if you think of Kibera outside Nairobi, people live in, you know, hundred or thousand dollar homes, you know, a piece of corrugated tin compared to our, you know, half-million, million dollar homes. So it鈥檚 wildly different, right? But if you think of their phones, it runs basically the same operating system that we run, it鈥檚 the same apps. It鈥檚 like barely any different. And so if we can figure out software based AI teaching that really does all the work, we can share that with the entire world. And so the project that CJ鈥檚 leading is trying to figure out one tablet per child in Rwanda, which is a great test lab. If that works as we hope, we鈥檒l do the hardware and operating system level, and various application developers in the U.S. will do amazing work there.

We鈥檒l put those together, and we鈥檒l see Rwanda rise to be the most successful education state, first in Africa, maybe in the world. And that will then prove at that point, which is the formula is really one tablet per child around the world.

Diane Tavenner: No pressure, CJ, no pressure. Number one in the world.

Michael Horn: We鈥檙e going to get all these people you鈥檙e working with, lots of attention out of this and so that we can multiply these efforts. Live from the ASU GSV Summit. Thank you, Reed, for joining us on Class Disrupted.

Disclosure: Reed Hastings was a founding board member of The City Fund, which provides financial support to 社区黑料.

This episode is sponsored by LearnerStudio.

In the study, , two unnamed school districts carved out dedicated time for hundreds of elementary school students to work with a well-known AI reading tutor, either during class time or after school. Researchers followed about 350 students across two randomized controlled trials. All of the students were expected to log on for at least two 30-minute sessions a week.

They found that of the students assigned to work independently with the AI, just over 60% in the first district and 53% in the second ever logged on to the platform 鈥� at all.

Among all students, average weekly usage came to just over two minutes in District A and just over five minutes in District B.

Those who did log on averaged 13.2 minutes a week in District A and 25.8 minutes in District B, using the tutoring for just four to five weeks on average in an 鈥渋ntervention window鈥� that ran from 14 to 31 weeks.

For Carly Robinson, the paper鈥檚 lead author and research director for the , the gap between access and use isn鈥檛 a shock. “As we’re talking about bringing AI tools into the classroom, the challenge isn’t just building good AI tools,鈥� she said. 鈥淚t’s getting students to use them and engage with them effectively.鈥� 

That’s going to take 鈥渋ntentional design鈥� that appeals to both students and their teachers, who must choose whether to offer access.

鈥淗aving these tools available, even if they’re really good, doesn’t necessarily mean they’re going to get used if they’re not being embedded into kids’ learning experiences,鈥� Robinson said in an interview.

But she was careful to note that the study didn鈥檛 draw conclusions about AI鈥檚 effectiveness, or the degree to which students were interested or uninterested in the bot, saying many factors could be at play. 鈥淭his is not necessarily the students not engaging,鈥� she said. In the two districts, the AI platform 鈥渨as likely one of many tools available to teachers.鈥�

For the study, researchers randomly assigned a group of students to work on the platform alongside a few classmates and a human tutor whose job was to support their engagement and motivation and to troubleshoot any problems students might encounter. In District B, the tutors were actually middle-school students who 鈥渉ad a free intervention block in their school day.鈥� A typical session included a short check-in, 15 minutes on the platform and a few minutes of reflection.

Pairing students with a tutor worked, Robinson said 鈥� to a point. Usage increased by roughly one minute a week in District A and 4.4 minutes in District B. The number of stories students completed each week jumped 71% in District A and 80% in District B. 

What the human pairing didn’t do was move the needle on reading scores: Neither district saw a statistically significant improvement in end-of-year reading achievement. But Robinson said the study wasn鈥檛 primarily focused on that. Rather it was looking at the overall impact of adding a human into the equation, someone who provides 鈥渁ccountability, motivation and relationship building.鈥�

Wednesday鈥檚 findings mirror recent ones from Khan Academy founder Sal Khan, who that the rollout of his in 2023 was 鈥渁 non-event鈥� for many students. 鈥淭hey just didn鈥檛 use it much.鈥�

Khan said AI tutoring doesn鈥檛 necessarily make students motivated to learn, or to fill in gaps in their knowledge needed to ask questions.

The new data also raise an uncomfortable question for educators: Among students who used the platform on their own, those who logged on tended to be higher-achieving and less likely to receive special education services. So the students who stood to benefit most from extra reading practice were among the least likely to get it. 

Robinson said she sees that as a red flag for anyone considering AI tutoring as a quick fix for underserved students: 鈥淚 think it should give us pause about treating AI tutoring as an equity solution.”

Alex Sarlin, founder of the newsletter and a veteran industry watcher, said the new study 鈥渟hines a light on several of the most persistent challenges in ed tech implementation: low usage rates that don鈥檛 meet dosage recommendations, differential technology usage based on prior student achievement, leading to lower usage among the neediest students, and a faulty assumption that students will jump into new tools without structured guidance.鈥澨�

The researchers鈥� approach showcases a promising direction, he said, 鈥渁s it is increasingly clear that providing access to tooling is not nearly enough to drive usage, let alone outcomes.鈥�

Amanda Bickerstaff, co-founder and CEO of , which provides AI literacy training to teachers, said results like these aren鈥檛 all that surprising, given what we know about these tools.

All GenAI chatbots, she said, can make mistakes, lack important context about students and how they learn best, and can provide biased outputs. Her group has recommended keeping these tools out of the hands of students through second grade, 鈥渁nd only with significant human oversight and AI literacy training鈥� for students in grades three through five.

鈥淎t this stage, there has been little evidence that GenAI chatbot tutors meaningfully impact learning outcomes for students,鈥� she said, 鈥渙r that they are developmentally appropriate for students in elementary schools.鈥�

Robinson, the study鈥檚 lead author, said she sees the usage findings as part of a larger pattern playing out as schools adopt AI tools more broadly. Schools, she said, should consider offering students 鈥渄ifferent iterations of these things based on what they actually need 鈥� and that’s probably a more likely pathway to scale than just saying, ‘Let’s give everyone an AI tutor.’ 鈥�  

Historically, personalized instruction has depended almost entirely on human teachers, with the teacher-student relationship central to the experience. But advances in technology 鈥� most recently in AI 鈥� have changed this dynamic, Robinson and her colleagues write. Now, personalized instruction exists on what they term 鈥渁 spectrum of relational intensity,鈥� from a consistent one-on-one human tutor to a computer platform that students navigate alone. 

AI tutors may approximate human interactions, Robinson said, but students may still benefit from the care and companionship that humans provide. Logging on and sticking with something that might prove to be difficult, she said, is easier with a human in the mix. 鈥淭here is just this component of accountability that a human can provide, where it’s so easy to look away or check out of something when it gets hard when you’re dealing with a screen.鈥�

A researcher at , a nonprofit focused on human connection in the age of AI, Lee asked: Why not?

In a word, the girl said: Instagram.

鈥淚f you try to dance at prom, you’re going to look stupid at some point,鈥� Lee recalled her saying. Eventually someone will pull out a phone and you鈥檒l end up on someone鈥檚 feed, seen by 鈥渢he entire school鈥� with mortifying results. Better just to play it safe. 

鈥淓verybody just goes to prom to look cute,鈥� the girl explained, 鈥渢ake a picture for the 鈥榞ram, eat and leave.鈥�

For Lee, who has spent years studying human belonging, that exchange unlocked an important, if unspoken, part of why AI holds such appeal. 鈥淲e’ve created this set of conditions where young people don’t feel like they have permission to be their real, unfiltered selves,鈥� she said in an interview. So they turn to AI, which is programmed to affirm them at every step.

from Lee and her colleagues offer this insight among others, painting a detailed portrait of how young people use AI and why. They surveyed 2,383 people ages 13 to 24 across the U.S. and found that for nearly half of them, AI has already reshaped their relationships in ways that are largely flying under the radar of parents, teachers and policymakers.

Among the findings:

• Just 15% of young people are in relationships with 鈥減ersonified AI鈥� characters 鈥� but for about 45%, AI is already reshaping their real-life relationships;
• 53% of young people say they set clear boundaries with AI, using it alongside 鈥� not instead of 鈥� human support;
• 61% say parents rarely or never talk to them about AI, and 53% say the same about teachers;
• Youth from low-income households are three times less likely as others to engage with AI, but they report greater feeling: 21% feel lonely often or all the time, compared to 6% of high-income youth; 57% feel like a burden to others, compared to 42%; and only 34% feel a strong sense of belonging at school, compared to 62%.

For the study, researchers sorted respondents into four broad clusters. About 28% rarely or never use AI, often out of ethical reasons or just disinterest. The largest group, 39%, uses AI primarily as a practical tool. They turn to chatbots such as Claude, ChatGPT and Google鈥檚 Gemini for homework and research, while keeping clear boundaries between AI and their emotional lives. 

Another 18% use AI for personal and relational support, such as venting about a tough day, seeking relationship advice and processing emotions. And 15% engage with AI characters and personas in more intimate, companion-like ways.

Within the four groups, researchers found nine variations that challenge the conventional wisdom around AI use. For instance, among those who use AI for emotional support were two very different groups. Rithm calls them 鈥淪ocial Processors鈥� and 鈥淧rivate Processors.鈥� While they may look similar from the outside 鈥� both say they have lots of friends and use AI to work through their emotions 鈥� surveys found that the Social Processors use AI as just one tool among many. The Private Processors, by contrast, use it as a substitute for real human interactions because they feel they can’t bring problems to those around them.

That data point could hold the key to understanding problematic AI use, Lee and her colleagues said, challenging the idea that lonely teens with small social circles are most at risk of unhealthy AI dependence. The data suggest something else altogether, said Kashyap Rajesh, a rising junior at Cornell University who consulted on the report.

鈥淭he driver of risky AI use is not necessarily isolation,鈥� he said. 鈥淚t’s feeling like a burden [to others] 鈥� and that came through in the research.鈥� 

The number of friends a young person has, the size of their social circle, how busy they are, whether they鈥檝e got family nearby and even their feelings of loneliness barely predict whether they鈥檒l fall into dependent AI use, he said. 鈥淲hat actually predicts it is specific feelings: Feeling like a burden to others, feeling like you can’t be your real self, feeling like there’s no one to turn to.鈥�

Julia Freeland Fisher, a researcher at the Clayton Christensen Institute who advised on the study, said that finding should help start a different kind of conversation around AI. 鈥淏urdening one another is building reciprocity, which is how we maintain the social contract, how we maintain social cohesion,鈥� she said. That young people are increasingly bypassing this step should be alarming, she said.

鈥淎I companions wouldn’t be nearly so disruptive to human connection if we had a sturdier social fabric,鈥� said Fisher. 鈥淚t’s the weakness of our social fabric that makes these [findings] so worrisome, not necessarily the technology itself.鈥�

鈥業t just keeps feeling easier than the alternative鈥�

For Lee, the finding on being a burden reframes so much of our understanding about young people鈥檚 relationship to AI. Virtually every survey respondent reported a specific 鈥渞elational rupture鈥� or crisis that made them turn to the technology. 

One young woman’s first question to a chatbot was, “I didn’t get asked to Homecoming 鈥� am I unlovable?” Another: “I got into a huge fight with my best friend, and I don’t want to tell anybody else because I don’t want them to take sides, so I needed to ask AI.”

“Story after story after story,” Lee recalled, “of a very singular, acute, discrete moment when they really had a moment of need and needed somewhere to put it.”

Rajesh, the Cornell student, said the data reveal a steady shift in which perhaps millions of young people are quietly moving from letting AI help with homework to asking it to mediate their emotional lives.

鈥淭hey start off using it to help them write an essay, or help them prepare for their interview, or to study for an exam,鈥� he said. 鈥淎nd they’re like, ‘OK, damn, this is really good, this is really helpful.’ And eventually their interactions escalate.鈥�

The drift happens gradually, he said. AI helps draft an email or respond to a text. Next it鈥檚 helping to navigate a social situation. Before long it鈥檚 processing a breakup.

Rajesh, who鈥檚 studying information science and AI policy, said his own AI use crept up on him: He went from studying with Claude to creating personalized AI study guides to wondering if even attending class mattered. 

鈥淚 found that how many times I go to class and how actively I’m paying attention in class is actually not the biggest indicator of my understanding of the content or exam performance,鈥� he said. 鈥淚t’s actually just how much time I spend with Claude dissecting the lecture slides and building study guides that work for me.鈥�

The report notes that because even productivity-focused platforms like ChatGPT, Gemini and Claude are engineered to interact with warmth and reassurance, what starts out as homework help or playful experimentation can evolve into a substitute for human interaction.

What adults get wrong

Alongside the findings on AI use, researchers found that how adults talk about AI is also potentially problematic: Their conversations are almost always about academic integrity 鈥� cheating, plagiarism, source citation 鈥� and rarely about relationships.

Rajesh said adults should be asking directly whether young people are using AI to process emotions, to rehearse hard conversations and to get support when they鈥檙e struggling. 鈥淭hose are questions that signal to a young person that the adult knows this dimension exists and isn’t going to freak out about it 鈥� which is, I think, the prerequisite for any honest conversation happening at all.鈥�

Michelle Culver, the Rithm Project鈥檚 founder and a co-author of the report, said young people tell researchers that when the topic is AI use, they’re 鈥渘avigating it alone.鈥� She suggested that adults approach the topic with 鈥渃uriosity鈥� rather than 鈥渏udgment or shaming.鈥� That could help both sides gain insight into each others鈥� struggles in the face of a technology that鈥檚 constantly challenging their reality.

In the same way that educators are worried that young people aren’t engaging in the 鈥減roductive struggle鈥� of learning academic content, Culver said, 鈥淲e similarly worry that young people might offload the relational work to AI and become ill-equipped to handle the very messy human friction of real relationships.鈥�

With many students and educators already using widely available artificial intelligence tools, state lawmakers and school districts are playing catch-up on AI policies.

In Maryland, for example, AI usage policies for K-12 schools are 鈥渁ll over the map,鈥� Democratic state Sen. Katie Fry Hester said.

In some school districts, she said, AI use is encouraged, while in others it is restricted, or 鈥斕齛 worst-case scenario for Hester 鈥� there is little to no policy guidance at all.

鈥淲hat we heard repeatedly is that the teachers were feeling like they had to navigate artificial intelligence entirely on their own,鈥� Hester said.

Hester said square one for lawmakers is AI literacy, which was the aim of new legislation that she sponsored and that was signed into law in May. It requires an AI coordinator in each school system, a statewide AI professional development for teachers and AI literacy to be a component of career readiness and computer science standards for K-12 students. It also requires the state Department of Education to provide certain guidance on AI.

Many other states have also been trying to create AI policies for schools. Lawmakers filed more than 134 bills across 31 states this year related to AI in education, focusing on data privacy, usage restriction in the classroom, literacy and training, according to MultiState, a government relations firm.

A survey by the Center for Democracy & Technology showed that (85%) reported using AI in their classroom during the 2024-25 school year, while 86% of students said they鈥檇 used AI for either personal or school-related reasons. But only about half of teachers and students reported that they received some training or information about AI from someone at their school, and few received training or information on risks of AI use.

A turning point for schools came with the rollout of ChatGPT in 2022, said Noelle Ellerson Ng, chief advocacy and governance officer for the School Superintendents Association. 鈥淎I was something that could not be gatekept,鈥� said Ellerson Ng. 鈥淚t was in the classroom the minute students were able to access it.鈥�

Her association does not take positions on state AI bills or policies. But she said districts are trying to avoid knee-jerk, reactive policies such as New York City鈥檚 brief 2022 ban of ChatGPT because of fears about cheating.

Some states have made progress in laying the groundwork for AI policy in K-12.

Ohio has set a July 1 deadline for every school district, community school and STEM school to adopt an AI use policy. The state鈥檚 model policy recommends that districts address student and staff uses, privacy, ethical use, teacher-specific uses, vendor agreements, third-party AI tools and student assessments.

A new signed in March requires local school districts and charter schools to devise local policies for AI usage in K-12 schools, requires state standards for AI literacy and education training and ensures that no AI 鈥渞eplaces or eliminates a human teacher.鈥�

enacted last month requires AI tools to be age-appropriate and requires teachers to review anything AI produces before using it in the classroom. It also allows parents to opt their children out of using AI tools. The law also directs the state education department to develop AI guidance and requires local school boards to set policies before the 2027-28 school year.

Yet even as schools are being sold on AI products by numerous vendors, there鈥檚 a growing skepticism about AI in classrooms. It follows a similar backlash about social media and digital technology鈥檚 academic and mental health effects on students, which has led to more states and districts putting in place bans and rethinking their reliance on laptops.

In the Center for Democracy & Technology survey, half of students said using AI in class made them feel less connected to their teachers, and 70% of teachers said they were concerned that students鈥� use of AI was preventing them from learning important skills.

Schools need to weigh the benefits of adopting AI tools in the classroom against their effect on student privacy, mental health and social skills, said Sue Thotz, director of outreach for Common Sense Media, a nonprofit advocacy group focused on technology and its effect on children and families.

Schools, Thotz said, may be the 鈥渙nly mandated safe space鈥� where students can learn to use and access emerging technology. But she and other education experts believe districts need to increase scrutiny of products.

Globally, the market for AI products in K-12 schools was worth around $391.2 million in 2024, and could rise to more than $9 billion by 2034, , a market research company. That includes AI products for tutoring, personalized learning, automated grading, lesson planning and administrative tasks.

鈥淲hen I talk about AI literacy, it鈥檚 not how to use AI. It鈥檚 understanding how AI is built,鈥� said Thotz. 鈥淲hy is it being created? Who鈥檚 profiting off of this?鈥�

鈥楪iving a tool to children鈥�

New York Assemblymember Robert Carroll said he uses artificial intelligence in his own work and sees its value. As someone who struggled with dyslexia as a child, he also thinks technology can help students with disabilities.

But he also wants to keep AI out of most K-8 classroom instruction. Students should learn basic subject matter first 鈥� in conjunction with critical thinking 鈥� and then later use the tools that can assist them, he said.

Carroll, a Democrat, has that would prohibit the use of most AI in K-8 classrooms, with exceptions for diagnostic testing and support for students with disabilities.

鈥淚t is imperative that all children gain strong foundational skills, especially in literacy and numeracy, and it seems that AI is uniquely positioned to possibly undermine that,鈥� he said. 鈥淭here鈥檚 a difference between giving a tool to adults and giving a tool to children who have yet to master skills.鈥�

Rather than full bans, most bills seeking to restrict AI have opted to focus on age restrictions, parental opt-outs, oversight and bans on using AI to replace teachers.

This year, Florida鈥檚 would have included a statewide restriction on student access to AI instructional tools before sixth grade, with exceptions for use supervised by school personnel, English-learner translation support and disability accommodations. It overwhelmingly passed the Senate 37-1, but died in the House.

A adds computer science to the required public school curriculum, including AI and emerging technologies. Connecticut lawmakers in 2025 failed to pass aiming to stop AI from 鈥渞eplacing鈥� public school educators.

Sophia Romee, the general manager of the GenAI Studio, an initiative studying how students and educators use generative AI at the College Board, the nonprofit that administers the Advanced Placement curriculum and SAT tests for high schools, said she is concerned that only that allow students to use generative AI have a formal policy governing its use.

The College Board鈥檚 research, Romee said, shows many students are worried about becoming too reliant on AI, and that adults need to give clearer guidance about where using AI tools for brainstorming, revising and tutoring crosses the ethical line into cheating.

鈥淪tudents are far more self-aware about AI鈥檚 risks than headlines suggest.鈥�

Like aviation in 1905

Jason Coley, director of the Center for Academic Innovation at Maria College in Albany, New York, said the policy debate needs to move beyond whether schools are 鈥渇or鈥� or 鈥渁gainst鈥� the use of AI.

鈥淭he better question is what kinds of AI use are supervised, age appropriate, transparent, and tied to real learning,鈥� Coley said. Schools need guardrails around privacy, student data, bias, teacher training and equity of access, he said, but also permission to 鈥渆xperiment responsibly.鈥�

Ellerson Ng, of the School Superintendents Association, said superintendents see AI as part of a larger umbrella of disruptive technologies in schools that has evolved from calculators to laptops to cellphones. The lesson, she said, is that overreactive policy rarely works. She also said schools should not cover AI in a separate policy, but as part of a broader technology policy.

鈥淚 don鈥檛 have a calculator policy. Why would I have an AI policy?鈥� she said, describing how some district leaders think about the issue. 鈥淚 have a technology policy.鈥�

With past technologies such as cellphones and laptops, adults could often control when students had access, Ellerson Ng said. With AI apps and platforms, many students accessed the tools before teachers, principals or state officials were even aware of them.

That makes bans difficult, she said. Schools can block tools on school-owned devices and networks, but 鈥測ou鈥檙e only one personal device away from social media and AI being in your schools.鈥�

Justin Reich, an associate professor of digital media at MIT, said that uncertainty around AI should make policymakers cautious about declaring best practices too soon.

Reich said states are trying to regulate classroom AI at a moment when the field is still so unstable that 鈥渨riting a guide for AI in 2026 is like writing a guide for aviation in 1905鈥� before airlines, airports or even commercial flight.

鈥淚f you were to take any of the AI literacy documents, AI readiness documents, even the moratorium documents, and put them against a checklist,鈥� said Reich, 鈥渢here would be a lot of boxes in the 鈥榳e鈥檙e making this up鈥� column and not a lot in the 鈥榳e have evidence鈥� column.鈥�

State lawmakers and school districts should be honest that they don鈥檛 know what they鈥檙e doing, are relying on limited expert information and that policy is subject to change with new information, Reich said.

鈥淟awmakers will need to be honest that what they propose now could be completely outdated in two years.鈥�

is part of States Newsroom, a nonprofit news network supported by grants and a coalition of donors as a 501c(3) public charity. Stateline maintains editorial independence. Contact Editor Scott S. Greenberger for questions: info@stateline.org.

I have an unusual window into what鈥檚 happening in classrooms as CEO of DonorsChoose, which provides resources in 90% of U.S. public schools. Each year, 200,000 teachers post requests on our site.

Since the 2022鈥�23 school year, requests for AI-related tools have surged more than 200%. But what鈥檚 interesting isn鈥檛 the growth. It鈥檚 the purpose.

Teachers are asking, overwhelmingly, for AI-enabled tools to reach students who have been left behind for decades: kids with disabilities as well as those learning English. In fact, 86% of requests are aimed at meeting the needs of students who have historically been underserved. In other words, teachers are turning to AI not only to save themselves time (although it can do that); nearly 9 in 10 are using it to get essential tools to the students who need them most.

For example, a middle school teacher near Atlanta requested AI-powered translation pens. With a simple scan, students can hear text read aloud or translated into more than 100 languages. For children who are learning English, or who struggle with reading comprehension, a $90 pen transforms their school day from frustrating to fulfilling. DonorsChoose has provided hundreds of these pens to teachers, along with more than 1,500 translation devices of other types.

In Chicago, an elementary school STEM teacher looked to AI to modify classroom materials when a child isn鈥檛 understanding a lesson.

In Miami, a middle school math teacher requested software that responds to students鈥� answers with immediate feedback that builds confidence rather than deflating it. Meanwhile, at another Miami middle school, a computer science teacher helps students get under the hood of machine learning by training robots to recognize and react to images. The project opens up discussions about ethics, real-world applications and how AI depends on what humans feed it.

In Detroit, high school educator Carrie Russell uses AI tools to effectively give every student a personalized tutor, expanding her capacity to teach each learner. She鈥檚 also mentoring other teachers about how to ethically and confidently incorporate AI tools into student learning.

These teachers aren鈥檛 asking for anti-cheating software or ways to monitor screen time, which is where much of the public debate is focused. They are experimenting and adapting tools that work for themselves and their students, without waiting for top-down guidance.

It shouldn鈥檛 be surprising that teachers are forging ahead and deploying AI in practical ways without directives from their schools and districts. Teachers have always been first responders to children鈥檚 needs.

In 2011, when American education underwent a seismic shift with states鈥� introduction of new academic standards, classroom teachers sounded the alarm on poor curriculum quality and misalignment to the new standards. Instead of waiting for the market or policy to catch up, they created materials that met the higher bar 鈥� and shared them with peers. 

More recently, on DonorsChoose, educators flagged the COVID pandemic鈥檚 effects on student mental health long before they became a national concern. We saw teachers request food for hungry students when SNAP benefits were disrupted last fall. And we routinely see teachers mobilize following natural disasters to replace what鈥檚 suddenly gone from their classrooms and restore some normalcy in their communities.

AI is the latest disrupter in education. It’s an opportunity to move toward a future when technology expands human potential rather than replaces it, where fairness is built into the design and where every student can experience moments of joy, discovery and magic. Teachers are showing what that can look like 鈥� one classroom at a time.

MagicSchool has emerged as a breakout artificial intelligence tool in education, with millions of teachers rapidly adopting it. But what鈥檚 behind that growth? How exactly does it use AI? And what does its adoption mean for teaching and learning? Its founder, Adeel Khan, joined Class Disrupted hosts Michael Horn and Diane Tavenner to go beyond the hype. Tavenner and Horn pushed for answers on the MagicSchool鈥檚 quality, privacy, and whether tools like this actually improve outcomes for students.

Listen to the episode below. A full transcript follows.

Diane Tavenner: Hey, Michael.

Michael Horn: Hey, Diane. It is good to see you. As always, it鈥檚 been fun because we鈥檝e had this arc of really digging deep on AI with school models. And then we鈥檝e been moving into the edtech tools that are starting to define a lot of current models, the new models we might see and so forth. And we鈥檝e got another big one. Dare I say, for today, this is going to be a good conversation.

Diane Tavenner: I am very much looking forward to it and looking forward to digging in. But before we do that, we have a quick ask for all of our listeners. Will you all rate or review Class Disrupted, wherever you鈥檙e listening to it? And of course, please subscribe. We鈥檝e never asked anyone to do this in seven seasons. And you know, this is 鈥� it turns out, this matters. So we鈥檇 really appreciate it if you would do that.

Michael Horn: Yes, we would appreciate it. And we know we get a lot of feedback from listeners. We certainly get a lot of emails, and texts and things of that nature. So we know y鈥檃ll are listening, but we need to see it in the ratings, reviews and subscriptions as well. It helps other people find the information, too. As folks know, this is a passion project for us, and we want it to actually be influencing the conversation more broadly. So, look, we know we got a lot of folks tuning in from our great partners with 社区黑料, Substack, Apple, Spotify, YouTube, you name it, we know you鈥檙e out there.

But subscribe, rate us, leave us a comment, and we鈥檒l just be appreciative for that.

Diane Tavenner: We will. And we will continue to be grateful for the direct feedback as well, because we use that. We take that in. And in fact, this conversation today is something folks have asked for. So we will not waste any more time.

Michael Horn: Yeah, no, it鈥檚 true. Well, I鈥檓 thrilled because we get to welcome Adeel Khan to the show, known obviously for MagicSchool. But before that 鈥� we鈥檒l talk about MagicSchool in a moment 鈥� but Adeel worked as a teacher, assistant principal, founding principal for DSST, the Conservatory Green High School in Denver in 2017, and then left a few years 鈥� I think it was like three or four years in or something like that 鈥� to coach school heads. And my sense of 鈥� Adeel can correct me if I鈥檓 wrong on this 鈥� but my sense is, this wasn鈥檛 the calling.

A year later, I think it was, ChatGPT burst on the scene into people鈥檚 consciousness in November of 2022. And Adeel says, there鈥檚 something here. This can revolutionize the teaching experience and you know, quickly sees like, hey, teachers aren鈥檛 going to figure out how to use this productively. Let鈥檚 help them do that. Builds MagicSchool and it takes off like a rock rocket. No exaggeration. I think over 7 million teachers now use it. That number鈥檚 probably out of date.

As I say. It partnered with more than 10,000 schools and districts, I believe over 160 countries. So we鈥檝e got a lot to dig in here. The use cases are broad. But Adeel first, welcome. It is great to see you again.

Adeel Khan: Thanks Michael. It鈥檚 great to see you too. And Diane, pumped to be on the podcast today, and thanks for the kind introduction. Mostly accurate.

Michael Horn: What did I miss? Yeah, what did I miss?

Adeel Khan: You got, you got it mostly there. I think certainly when I got to coach principals, I didn鈥檛, it wasn鈥檛 my favorite thing in the world. Not because it wasn鈥檛 a great opportunity. I worked on a great team, and principals need support. But rather, I think, when I was founding my school and building it, I thought, you know, if you鈥檇 have asked me if I had the most important job in the world, I would have answered yes. And I just missed that feeling. I didn鈥檛 have that kind of builder-chasing, really ambitious feeling when I was in that new role. And it鈥檚 sometimes good to know that about yourself, right.

I think that like sometimes trying something and knowing that you know where you are because, you know, in some sense at the end of my principal career and seeing the graduates in the first class of seniors, as I kind of built the full school out, I was exhausted. Right. And sometimes you think about, you know, what鈥檚 the other side like, you know, and you know, being an advisor sounds less intense than being in a school building every day and you know, chase this big dream and, and then, you know, knowing what the other side was like, I was like, oh no, some people are wired for this. Right? Being in the builder seat and being… I learned that about myself. And even when things get hard today, and they do, we have a lot of challenges. I know that this is what I want to be doing and, and I would be, you know, I wouldn鈥檛 be as… there鈥檚 no grasses greener thoughts anymore. Even when those seep in, I鈥檓 like, nope. I know this is the thing.

Even when it is, I鈥檓 in my most trustful habit.

Michael Horn: So I鈥檓 really, I love it because you鈥檙e willing to make the trade offs for the thing you really want. And obviously we hope all our young people start to grow that same muscle, right, in their lives.

Diane Tavenner: That鈥檚 the work I鈥檓 doing right now. So it鈥檚 an inspiring story.

Michael Horn: Well, I was going to say. Right, so like we have two, you know, Diane, Adeel, you know, founding school heads become founding ed tech company CEOs. I think the story of the founding story of MagicSchool is somewhat well known, but maybe what鈥檚 less known, Adeel, is how you think about the problem that you鈥檙e actually solving both originally and today because you all have expanded your scope quite a bit. But let鈥檚 start at the outset itself because I think your framing tells us a lot about why you grew so quickly. So, you know, how do you describe what MagicSchool first solved for teachers and why that approach was so important at the time it came out?

Challenges using general AI tools in education

Adeel Khan: Yeah, so I think that there鈥檚 a part where there鈥檚 maybe a little revision on the way you describe the problem that I found, which was not that I didn鈥檛 think teachers would be capable of figuring out how to use this and use it productively, it was that the tool itself was a general purpose tool. It wasn鈥檛 built specifically for the domain and therefore it was clunky to use. And you know, at that time even really good prompting didn鈥檛 yield a really good result for teachers. It feels like, you know, ancient history that you had to have massive prompts for like an LLM to understand what you really meant in the prompt. But at that time it really was, so if you were prompting ChatGPT, it was like you could get something quality, but your prompt had to be a page long for it to really understand your context, really understand the format of your output, just all those different things. And if teachers were going to use it right away, we had to do some of that work to get them to a magic moment with the technology. A lot of teachers would report to me when I was training them on ChatGPT, which is kind of the way the product idea started was, you know, I was prompting it so much, it wasn鈥檛 saving me time, it was costing me time.

So there was like kind of a really acute prompt of, you know, promise that this has an opportunity to assist you, save you time and also make you more efficient, augment your abilities. But it鈥檚 not working for you right now or the vast majority of teachers in that moment. But I think of it as, you know, when I started thinking about building the product, it started with I actually thought I might be training teachers on ChatGPT, and that just didn鈥檛 work. I like, experimented with it. I went to my school and teachers didn鈥檛 really adhere to the product. So I was like, there鈥檚 got to be a better answer to this. I was savvy enough to kind of be looking around to see, like, you know, products that were coming out that were generative AI products. And they were like this first set of products that were hitting the market.

There was like Harvey, which was like AI for lawyers. I remember that really vividly. And I don鈥檛 think they even had a live product, but it was like announced as a product, and they鈥檇 gotten some funding, and there were news stories about it. And I was like, oh yeah, there鈥檚 going to be an AI that is made for each kind of profession. And that makes sense because each profession has its own quirks. It has its own domain knowledge, it has its own workflows, it has all these things that are specific to that industry. And a generalized alternative language model isn鈥檛 going to be the answer for every single one. So that was kind of the clear vision from the start, is we make a really domain-specific, intimate product for teachers.

And the products evolved a ton since then, but that core of it was like, it鈥檚 going to be really domain specific. Even in those early days, I thought to myself, we鈥檙e going to make some tools in this platform that if you鈥檝e never worked in a K12 school, you would have no idea what that even means. Like, that was kind of my bar for myself is like, I want it to be so intimate, so clearly made for teachers that they scroll through that initial dashboard of tools and they see a couple things and they say, you know what? Somebody who鈥檚 worked in schools like, built this. I still hear that today. And it鈥檚 one of the things that brings me the most pride. Teachers don鈥檛 necessarily get built products that are so clearly intimate for them and know their work. And that鈥檚 what we hope to do and continue to do with the platform.

Exploring MagicSchool鈥檚 AI features

Diane Tavenner: This is such a good place to jump in because one of the things Michael and I are doing this season is trying to help people see how AI is truly being used in schools like today. And so building off that origin story and where you started and where it is today, I mean, let鈥檚 get into like how MagicSchool actually works and what, where the AI is in it. And, you know, this is the moment we get to kind of nerd out, which is fun. It sounds like you鈥檙e, you鈥檙e in that space too. Like, just to give people some context: When you enter as a teacher, it seems like there鈥檚 basically three big categories of support, teacher tools, writing feedback, and Raina the chat bot. And so, and then when I look at the list of teacher tools, you鈥檙e right, I open that up, I鈥檓 like, oh my gosh, there鈥檚 like 75ish tools. And we鈥檙e getting everything from like a worksheet generator to report card comments, unit plan generator, standards unpacker, IEP generator, quote of the day, classroom management plans and even tongue twister, you know, so like help us unpack.

Like literally what鈥檚 going on here in this tools section? How鈥檚 AI? What is the role of AI and what, how are teachers using this?

Adeel Khan: So one thing that I think MagicSchool does exceptionally well as a product is, while there鈥檚 a breadth of tools and sometimes we get the feedback that it鈥檚 overwhelming, it鈥檚 like this tension we feel between it being overwhelming and also really familiar for teachers. Like, I鈥檝e gotten that feedback since the very start and I always hesitate to like change the UI or make some meaningful shifts there because it has kind of become our calling card that we鈥檙e this like list of teacher terms almost. Right. The UI itself has been replicated by 20 other products at this point because of work. Right. It鈥檚 like, you know when you say AI to maybe a skeptical teacher or a teacher who鈥檚 like, technology鈥檚 not actually been a meaningful part of their story in the classroom, or a teacher who鈥檚 maybe been sold on technology being a really big opportunity for them, and they鈥檝e been burned. And there鈥檚 all kinds of reasons why teachers and technology have not gotten along particularly well.

I will be clear. I was not somebody who like, as a principal, saw incredible value in technology. My school didn鈥檛 subscribe to any boxed curriculum-type technology tools or platforms. We built all of our own, internal. We built all of our own curriculum. We didn鈥檛 subscribe to a single ed tech product outside of the LMS that we used. And that was a kind of a requirement. And the reason was mostly because I just didn鈥檛 know that they were going to drive outcomes for our kids. And it was an outcomes-oriented principle.

I wanted to make sure our kids鈥� literacy and MAD scores and their SAT scores were like, that was, kind of our focus was preparing them for college. And I wasn鈥檛 sure that any of the tools out there would make a difference. And I might have been just wrong. Like, you know, maybe they just weren鈥檛 in my world. And there鈥檚 some great tools out there, I鈥檝e learned since joining the ed tech world. But nonetheless, I think there is that skepticism amongst folks, and there are tools that have come to schools and have done just about nothing. I mean, everyone who鈥檚 worked in a school district or building will tell you, like, remember two years ago, that initiative we had? What happened?

That is just like a common story, like a new curriculum that鈥檚 going to revolutionize the way that you teach and can completely, all those kids that are behind your class, they鈥檙e going to catch up in a year. Or here鈥檚 a new tech tool that is promising to do this. So there鈥檚 all this baggage around, like products that come into schools. And the baggage often is, I invested so much time and energy in implementing this thing, and I don鈥檛 know how much value I actually got out of it, and then the district gave it up in two years. Like, you know, like, there鈥檚 like frustration here. So if you think about MagicSchool coming in, like really novice, you know, founder 鈥� me, I was just like, let鈥檚 get people to value really quickly, right? Like, let鈥檚 just get them into what this thing can do so it鈥檚 believable.

Using generative AI in education

Adeel Khan: So you jump into MagicSchool, you see the rubric generator, you click on it, it asks you some pretty simple like form fill up type questions like, you know, what we鈥檙e going to do attach the document maybe of the assignment you鈥檙e going to be assessing with this rubric. All very simple, super easy to use. And then you click the generate button, and you get a rubric that looks like a rubric. It鈥檚 what you expect it to be. It works the way you want it to work. It meets your expectations. Whereas like at the, at the, you know, in early days, ChatGPT and even still today, like it鈥檚 even the simple friction of if you went into a regular chatbot and you tried to build a rubric, of course you could, but like, you鈥檇 have to type in like four sentences.

I want a rubric. It鈥檚 gotta be six columns. It鈥檚 gotta be, you know, like making sure a boxed format. And then like, you know, you do it and then it, then the LLM will respond something like, oh, did you, you forgot this detail? Can you give me this detail? Wait, wait, what? Like I just gave you, like, it鈥檚 a frustrating experience for somebody who was told this is gonna save them time. And I have to know funky things about prompting nonetheless, like, that鈥檚 delight. A rubric generator that gives you a rubric is like a zero time-to-value, incredible experience for the actual. I see how this is going to be really helpful to me in my daily work. And the flywheel it creates is also a secondary value because if generative AI can do this for me, then what else can it do for me? And we have a chatbot on the platform Raina that is just like a ChatGPT but built for education with a couple, you know, bells and whistles and education-focused context. And if you use that, like, if you know that, you know, MagicSchool could create a rubric for you.

No, we鈥檙e not hiding the ball here. Like, you know, what鈥檚 happening is there鈥檚 a prompt, there鈥檚 inputs, it鈥檚 coming out with something. You know, you can probably prompt Reina and do something completely original or new, but you need to be able to see the value first for you to be really sucked into it. And like, I think an amazing job of getting people in the door with something really low stakes that should like, turns out to be really high quality, saves you a bunch of time and makes you want to use the platform more. It makes you want to think about, okay, what are the other tools? If it did this thing for me, what else can it do for me? And I think that鈥檚 the way that we see the flywheel start. We think about the user.

They start with these really simple tools. They might graduate into their own free form prompts and Raina and trying things. And then they might decide to use an agentic workflow like our class writing feedback tool, connect their Google Drive and have writing feedback dropped on each of their documents. Then they might want to think about, well, if I鈥檓 using this in my work and it鈥檚 really valuable, it鈥檚 helping be a thought partner to me. I wonder if it would be helpful for my students typically. So then we 鈥�

Diane Tavenner: Yeah, super helpful to understand your thinking and your flow, and it seems like that is working. Let鈥檚 go back and nerd out a little bit on the rubric generator. So where…? What happens when they type that simple thing in? Like, how are you using AI literally? What鈥檚 going on behind the scenes there? And as a person who鈥檚 written a lot of rubrics in my life, used a lot of rubrics in my life, like, how do I know that鈥檚 a good rubric and, like, what鈥檚 going on behind the scenes and under the hood?

Adeel Khan: Yeah. So great question. When MagicSchool started, it was just a prompt. It was like me with a prompt in the background. I had created rubrics as an English teacher in my life, and the best instructions I could give, and the judge was me. Like I was whether or not the quality was high enough or, you know, would meet my standards. And then it was a group of users who were doing the same things upon using it.

Diane Tavenner: And you鈥檙e prompting one of the big LLMs, essentially.

Improving model quality and selection

Adeel Khan: That鈥檚 two and a half years ago. So just know that鈥檚 like, it鈥檚 radically, radically different. So today we have a trust, safety and quality team, running our own evaluations on the platform. We have pedagogical experts on the team who are reviewing poor quality. So there鈥檚 a human part of it. Then there鈥檚 a thing called LLMs as a judge. Some of our evaluations are using models like Claude to judge the quality of the rubric. I was actually on a panel yesterday with one of the, a teammate, one of the teammates at Anthropic, his name is Nirob, and he was, he was naming that, you know, Claude itself at this point is probably the level of like a PhD in domain.

So, you know, you can have Claude, a PhD in a domain, almost, nearly, and getting better every day, judge the quality of the rubric too, and then iterate on the prompt model we鈥檙e using and change those things out. So we actually have, we have a wildly more complex version of judging for quality internally now. And it鈥檚 a combination of like LLM as a judge, human in the loop, user feedback on the platform. And then when new models come out, we are, on a regular basis, once every week or two, we鈥檙e saying, is this new model better performing against our criteria? And so each one of those essentially roll up to a score, and we have a score that determines whether or not this model can exist in the platform. And sometimes, you know, if it鈥檚 a 94 from one model and a 95 from one model, but one model is like one tenth the cost, we will choose the one that鈥檚 a little bit cheaper because we have to keep our platform affordable for schools. The vast majority of them, we鈥檙e able to say the absolute best output is the one that we go with, but they鈥檙e judged across a lot of different quality markers now to ensure that the output is something that we can stand behind.

Diane Tavenner: So just so I make sure I鈥檓 understanding. When I go in there and I pick a teacher tool, like, I think it鈥檚 report card feedback or something like that, you all have kind of established what good report card 鈥� the elements of good report card feedback. And then you鈥檝e created prompts that are prompting the LLMs to do that. And then you鈥檝e gone through and tested those responses to make sure that their quality. Because literally the humans aren鈥檛 testing while I鈥檓 the teacher here asking for that. You know, it comes back like that. Right? So.

So you just have confidence in those responses I鈥檓 getting?

Adeel Khan: Yeah. So, yeah, we鈥檙e essentially running out, like, hundreds of queries against them and then judging queries and then orienting around what the best kind of set of all of those variables. What鈥檚 the best prompt? What鈥檚 the best model? What鈥檚 the best output?

Diane Tavenner: Okay, got it. Now help me understand, like, that rubric. So now, like, I鈥檓 a teacher in my classroom, it鈥檚 often been standard that the teacher, sort of in their classroom, doing their thing, using their own rubric. How do you think about it from, like, your principal seat, like, the whole school? So, I mean, you know, for example, when I was leading schools, we built a longitudinal rubric that, you know, over time, the kids were 鈥� and so one of the things, as I play with MagicSchool, I鈥檓 like, oh, what鈥檚 the bigger picture here? You know, what鈥檚 the 鈥� what鈥檚 the high school arc? Or what鈥檚 the whole arc? And is this kind of more of an activity in my classroom base? Or is it the full big picture, the backward planned approach? How do you think about those things?

Adeel Khan: Yeah, it鈥檚 a really great question right now. I think that MagicSchool鈥檚 teacher side is best judged as kind of an assistant that helps you in the moment that you need it in your daily classroom activities. You can use Reina as a thought partner when you鈥檙e struggling with something or a Chatbot, you could, you know, really build out, like, a full, like, unit, then the subsets of lessons and like a generative thread, all the tools are threaded. So, for example, if you started with the unit plane generator, you could then say, build me the first lesson in this unit. It keeps context of that entire thread, and you could give it, like, input about, hey, my students performed this way on it. Can you generate the next lesson with that context? And you could do it that way.

But transparently, few people do. Right? Like, it鈥檚 really like, they come in, they kind of get what they need, and they keep going, we want to move toward more of what you鈥檙e describing, which is like, hey, the entire cycle is brought through in the platform. You give the platform your intent, and you keep these really rich threads with memory, context and knowledge of your classroom. So, we鈥檙e moving toward where you can kind of have the entire context of the classroom experience built in the platform. One of the key components of that is assessment.

Building personalized learning assessments

Adeel Khan: Like if the platform understands how your students did on their assessment, say for example today you could build an assignment in MagicSchool and you could, you could actually, you could build an assessment in MagicSchool. It could be taken by your students in MagicSchool. The results of that assessment could produce some really interesting insights about how they performed against the standard. And you could then create a material based on that assessment. So that鈥檚 where we think it鈥檚 going, is that hey, you鈥檙e going to build things based on the assessment results that take into context your students areas of strength and areas of growth based on the insights that are in aggregate across those folks who took the assessment. And you can imagine that there鈥檚 even more you can do, right? Like there鈥檚 on the student side, on an individual level, if you had assessment data about a child and you also had them taking assessments in the platform, you had them doing activities in the platform, you could trace a student鈥檚 personalized learning profile that understood kind of their academic needs. It could be continually updated with memory around the way that they interact with a platform. Not just from their academic strengths and how they鈥檙e growing, but also their stylistic preferences, their post secondary desires.

And you could really make these incredible persistent learning experiences for students that they always can tap into that鈥檚 associated with a really rich profile to serve them right in their zoning proximal development and in a style and in a way that they will engage and learn more.

Michael Horn: I鈥檓 curious off that just because that, that like becomes a very big vision, right, that shapes around the student how starting from like a teacher workflow, right to like the student life cycle almost, if you will, how much do you need to know and collect about individual students and how much do you need to know about the specific? Like take the assessment question, right, like, so I鈥檓 going to give feedback on you鈥檙e grappling with a particular poem or a book or something like that. How much does the model then need to get trained not just on the teacher rubric, but actually on the content itself as opposed to just the standard, which is probably a higher-level statement of ability to do X. But ability to do X in one context might be very different from another. So like, just help us think through the scope of that, of getting to that vision you just painted.

Adeel Khan: Yeah, it鈥檚 a really good question. I mean, we are at the early stages of understanding how this will work without student data. But as we build so one of the things that we are doing or in the planning process of getting some of these ideas off the ground, there is a really interesting process where you can kind of use LLM agents as sample students and then you can also assess the quality of their interactions with the platform based on specific profiles that you create for the AI agents. And then you can kind of get a really good data set on what you鈥檙e describing, Michael, is like, is it working? Is it actually meeting your needs? Or is that student who鈥檚 an agent who has the stylistic preference around visuals for their needs. That鈥檚 the way that they learn best. Or audio is the way they learn best. Their reading level or their reading test score diagnostic was this.

And this is what that means. You can kind of create all these profiles individually and then you can run them against the interactions they have with the platform and then you can judge how high quality the interaction is on the platform with like an external observer. So, like, almost similar to where I described that LLM as a judge, the judge can actually judge the experience.

So you can think of it as almost like a principal judge. Right? There鈥檚 a principal who鈥檚 an agent who鈥檚 watching the AI interacting with the student and determining is that a high quality interaction or not based on the child鈥檚 profile, based on what I know about what high quality instruction looks like. So you can start building these kinds of recursive loops and running them hundreds of times and get to some pretty, pretty impressive verifiable outcomes pretty quickly. And of course, you know, AI is not kids. And they鈥檙e not going to be perfect.

Michael Horn: AI is not going to act out for 鈥�

Adeel Khan: Yeah, yeah, yeah. So I don鈥檛 want to, you know, kind of anthropomorphize that this actually is the answer, that we can simulate everything through generative AI. But it鈥檚 a great way to like in a, in a basic sense feel like, okay, it works in a simulation. And then when we bring it to students and we work through the pilots and we see schools and districts who embrace this, then we can see it in action. We can combine those insights with the insights we got from our simulated experiences and make something that we think is really powerful. Some of those generative experiences, we鈥檙e going to be starting this summer in summer school with some partner districts to see how it works in practice and is it actually the needs of the kids. And we鈥檙e going to use that data as data to inform the product as it gets into more live cadence.

Diane Tavenner: You just said so many interesting things in there. Like, I am of like five minds right now. Where what thread do I want to pull? But let me pull on the one, because you started by saying, like, we鈥檙e playing with what we can do without student data in there. And best I can tell, based on my experience in MagicSchool you鈥檙e not connected to an SIS or anything like that. So you鈥檙e not pulling in any data as the teacher around your students. That said, I did create, I did use the IEP generator and I generated an IEP and I, you know, I, look, I didn鈥檛 do it on a real student, obviously. I鈥檝e got like 25 years of experience. So I sort of built a proposed persona in my head and like input the information.

Discussing AI-generated IEPs and privacy

Diane Tavenner: And I will say I was like, kind of blown away that a fully formed, detailed, and dare I say, very confident IEP like, came out. You know, and at first I was like, oh my gosh. And it looked like an IEP that you would kind of read in a school sort of thing as I was skimming it. Then I started like really digging it, and I was like, oh, as someone with 25 years of experience, like, this is not the IEP I would have written necessarily because I, in my mind, you know, the kids I鈥檓 thinking about, like, I have intimate details and I鈥檓 like, wait, that feels a little, it felt a little AI-ish, right? Like sometimes AI gives you really, like, seems like compelling results, but they鈥檙e not very specific or personalized or whatnot. And so that was one thing I was curious about. Like, how do you think about some of the tools that are like that and, and how they get used? And like, I was thinking I鈥檓 a first-year teacher, and I use that, and I don鈥檛 have 25 years of experience. Like, can I? How do I do that? And then the second piece is you鈥檙e out with teachers, like, do they just pour a lot of information about kids into the platform? And I鈥檓 sure you get a lot of questions about privacy and security and, like, what鈥檚 going on there?

So I鈥檓 curious about those two angles.

Adeel Khan: Yeah. So I mean, we do a lot of professional development. I鈥檒l start with the second one. We do a lot of professional development around making sure that teachers do not share information that could be sensitive into the platform. So there鈥檚 upfront training in that tool you used, actually, you鈥檇 see that like it actually actively reminds you in the tool itself to not, because that鈥檚 a tool that鈥檚 more susceptible for you to maybe submit accidentally that data. Of course, if any data is brought to our attention that was submitted to the platform, those PII, we remove it immediately.

And we have incredible data handling practices, and safe things are all available on our website. So we鈥檝e done a lot of things to make sure that schools and districts can trust us. And so that鈥檚 certainly something that we consider. And we just think training and enabling teachers is the best way to prevent that stuff. Because even if MagicSchool has really great data handling practices and is kept safe, like they might bring it to another platform and you know, we want them to know how to use our AI, but also any generative AI and keeping student data safe is incredibly important when you鈥檙e using these tools.

Diane Tavenner: Yeah.

Adeel Khan: Second part, Diane, I think is really interesting. So I was a special education teacher too. So for me, I can think back to when I was a novice teacher, and MagicSchool would have been a godsend for me. The IEP would have prevented me from being anxious. I probably would have helped me save a lot of late nights, and it would have made me feel like I had real strategies to support kids because, you know, it鈥檚 a good point. So, so your question, I think I might challenge this. Like you presented it almost as like a fear that they鈥檙e not 25 years, so maybe they鈥檙e just taking this robotic IEP that鈥檚 not as good as the high-quality, 25-year IEP. I want you to know that like I鈥檝e been a principal reviewing IEPs and I have seen teacher wonderful, hard-working teachers Submit IDPs with the wrong names in them because they were just copying goals and pasting goals from student to another because they were just trying to get it in by the deadline.

Right? Wonderful, hard-working, incredible teachers. So do I believe the world has gotten better because of MagicSchool鈥檚 IEP generator? Yes, it has. The floor has been raised because there are actual, because now the barrier is you understanding the student and getting some, some high-quality things and you know, even a novice teacher will see the goals and at least they have, they know how to write goals now.

I didn鈥檛 know how to write right, like, and I didn鈥檛 necessarily have someone to go to to help me write those goals. So I would say net MagicSchools raised the board dramatically for the way an IEP is written. I think a family would be much happier to see a MagicSchool-written IEP than the ones that I was editing, if I鈥檓 being totally honest. And it鈥檚 again, not because teachers aren鈥檛 wildly hardworking and talented. It鈥檚 because there is no time. And so I think that鈥檚 kind of the reality now. In an ideal circumstance, you know, they have a really great draft and they have an instructional coach like you, who they can go to and say, what do you think about these things? And like, you know, they can question and challenge them and push them and make them even higher quality.

But I do think that, like, sometimes we miss the reality of what happens in schools when we were critical of the tools that teachers are using to better their practice. And sometimes we just need to trust teachers. Like, in that case, I鈥檓 like, actually, I even trust the first- and second-year teacher to use this appropriately. And especially if they鈥檙e educated and they鈥檙e told, like, the way to use it, not to submit appropriately private information, things like that. But I always challenge people is like, yes, you know, people, we have schools and districts that hide the IEP generator because they鈥檙e so scared of it. And you know, we respect anyone鈥檚 decision on what they want to do and they鈥檙e allowed to do that and like, you know, our enterprise product and then we support them in doing that. But I, on a personal level, I always challenge them. I say, like, look, like, ask your assistant principals who are reviewing IEPs.

Yeah, they see, are they, are they higher quality because you took this tool away or are they higher quality when MagicSchool is in the loop? I think that鈥檚 like the question to ask is what鈥檚 the before and the after rather than like, what鈥檚 the ideal? Right. You want the ideal. But I think that there is like a, this go between. I think it is super strong. And I think that the tools are getting even better over time. And I think that the better the context that you share with it, the better it鈥檒l do of course.

Diane Tavenner: Yeah, that makes sense. And I think I understand the perspective that you鈥檙e coming from and certainly your lived experience. And I do think sometimes people, you know, have an idea of what is happening in school versus a reality of what is happening in school. And so I appreciate that. I was also really tuned into you saying that as a school leader, you all developed your own curriculum. That was my reality too. You know, we ended up building curriculum, we ended up building Summit Learning, which was like a whole massive curriculum. And so I鈥檓 wondering how.

But a lot of schools have adopted curriculum, as you know, how do you recommend that teachers sort of deal with the world of like, I have an HQIM that I鈥檓 given by my school and I want to use MagicSchool. How do those two things play together? Or do they, or what does that look like?

Adeel Khan: Yeah, I think that, you know, again, lived reality of an HQIM in a district. I鈥檝e lived in that reality as well. We didn鈥檛 box any curriculum, but sometimes we would get for a certain subject we would have like, you know, free prep lesson plans or whatever it might be. And I think the lived reality of those things are teachers are always modifying, changing, supplementing, making those things work for their classes. And that鈥檚 good. So I think that that鈥檚 what we鈥檝e encouraged them to do is like, yeah, keep the spirit of what your school wants you to do. And obviously there鈥檚 a research base behind the curriculum that you鈥檙e using, hopefully, and we want to be a great supplement to that. We want to make sure that you鈥檙e able to build the supplementary materials that you need to make sure your class functions and works.

Integrating curriculum with MagicSchool

Adeel Khan: And MagicSchool could be used alongside those things. There鈥檚 a world where we build knowledge into our platform so the schools and districts can upload things like standards, curriculum they鈥檝e built internally that are not like, you know, copyright by the publisher. And there鈥檚 a world where we partner with publishers too, and we bring those knowledge bases into our platform and call them as well. And as you鈥檇 imagine, those publishers aren鈥檛 super excited to work with large language models because this is like their proprietary IP. And nonetheless, I think that like, you know, we think that there鈥檚 a future where there鈥檚 kind of a win-win for both of us in a world where teachers are finding that they鈥檙e starting their day with MagicSchool, and it鈥檇 be really convenient if they can pull in some of that information and make those curricula more flexible with generative language models. But yeah, I support that. I鈥檓 also, I think a lot about curriculum and I make spicy posts on curriculum on my LinkedIn, if you haven鈥檛 seen them. There are a couple curriculums that are incredible and obviously they鈥檝e driven meaningful outcomes for kids.

And I don鈥檛 think they鈥檙e all that way. To be clear. I don鈥檛 think they鈥檙e all super research based. I think that a lot of them are not particularly valuable and I think, I don鈥檛 think. I know a lot of teachers hate being put in a box. They hate having to be told that you can鈥檛 be autonomous in your classroom. You must follow the script because the script is better than anything you would create.

Which is like the, not the intended message, but the unintended message that a teacher might receive when receiving certain curriculum. And of course they鈥檙e the teacher鈥檚 love. Right? Like they tell you, no, this thing鈥檚 amazing. It鈥檚 changed my classroom. So not painting with a broad brush. There鈥檚 also like really amazing ones. I will say that my lived experience has shared that like if you trust teachers to go find the right resources and give them the tools to do that and know their kids and coach them really well, you can get really extraordinary outcomes. Mind you, most of my experience is secondary, but we have incredible results for our kids.

We served a highness population and had dramatic growth. And so that鈥檚 my own experience. So I don鈥檛 know. I think that there鈥檚 a little bit of like, I would not call myself as a personal. Like on a personal level. I do not ascribe to the Church of Box curriculum. That is not my ministry as we used to call it in Atlanta.

So do I think there鈥檚 really good stuff out there? Absolutely. It鈥檚 not my ministry. I just teach it.

Diane Tavenner: That makes sense. Clearly you are outcomes driven. I know that based on the school that you started and all the language you鈥檙e using here. How do you think about 鈥� how do you now at MagicSchool think about what outcomes your 鈥� how do you hold 鈥� what鈥檚 your bar? What are your goals? Like what are your outcomes that you want MagicSchool to drive to and how do you measure them?

Adeel Khan: So there are a couple ways that we鈥檙e thinking about this. So outcomes are at the heart of our mission. We have named goals in our company about how we are going to drive student outcomes in classrooms. Right now, the way that looks in our platform is the amount of what we would call feedback delivered to students. So generally what we鈥檝e seen in the first two years of the product is that like the things that we, that teachers have said have driven growth, the experiments we鈥檝e set up or have not set up we鈥檝e just been reported are that when students get things like feedback on their writing that they鈥檝e done on their own aligned with rubric in the platform, that is powerful. That is something that drives an alchemy classroom.

Measuring feedback and student outcomes

Adeel Khan: We have actual classrooms that have shown state exam scores change over those things. We have enough data to say that like when MagicSchool gives student feedback that teachers is controlling and aligning to a rubric or the way that they鈥檒l assess students, that鈥檚 a great thing. So we measure right now how many instances of feedback is MagicSchool giving to a child under the supervision of their teacher through either our assessment platform in the product which gives students just in time feedback after they take an assessment, or in their feedback portion of the platform where they tune in. So those to us are pretty hard. Like we don鈥檛 kind of look at it as like hey, you talk to a chatbot, so you learned. Like, that鈥檚 not enough for us to kind of count and like our something that we think is definitely going to drive an outcome. It certainly might drive AI literacy and we certainly think about that measure as well. But in terms of outcomes, that鈥檚 the way we think about it now.

Well, where we want to go is we want to be able to probably say that students have learned and judge it in the platform itself. So one of the ways we might do that is through having a diagnostic assessment students taking the beginning of the year and then at the end of the year or tracking timing platform as we have more persistent student profiles in the platform and simply asking the district to themselves compare their users data which students spent the most time in platform and how much did they grow and then just give us a report back. There鈥檚 ways that we can kind of say it鈥檚 not a perfect correlation, but if students are spending more time in the platform and they鈥檙e getting better results at the end of the year, they鈥檙e growing more than their peers. That鈥檚 usually a pretty good indicator to us. There鈥檚 an experiment that was run unbeknownst to us in our first year. I actually shared this on a panel yesterday. I was at South by Southwest.

But I think it鈥檚 a powerful example and I think what a lot of organizations are doing around generative AI around the world and I actually think this is fun because a school district was ahead of enterprises, they鈥檙e ahead of companies in the way that they鈥檙e thinking about generative AI and our first year that we had a real enterprise product was 2024 to 2025 or 2023. Our first partner, which you imagine our very first partner at MagicSchool would be pretty innovative. And they certainly were as Aurora Public Schools, one of our very first ones. At the end of their first year using MagicSchool, they got a printout of their MGP scores at the district level. MGP in Colorado is basically like a growth score that is associated with each teacher in the district based on their state exam scores. Their grade has a high stakes exam. So the district quite literally gets a list of all the teachers who had the highest growth in order of like this calculated growth score. So they kind of have like which teachers are having the best results in their classroom.

It鈥檚 pretty sophisticated calculation. It tracks like based what was their expected growth based on their prior year鈥檚 performance to this year鈥檚 performance. It鈥檚 a pretty solid number. Like at the end of the day it鈥檚 like the kids actually grew and, and it鈥檚 based on some pretty hard metrics. What they did was that they liked the way the story is told, is the academic, one of the assistant superintendents called the technology director and said, I have all the teachers who have the highest scores or the highest MGP scores in the district. Can you pull up MagicSchools our user dashboard which shows which teachers are using it the most? And they said like one for one. It was like best growth scores were in the top 20 users, best growth scores, best users.

Using AI to boost engineering productivity

Adeel Khan: So the way industry is doing this now, what we鈥檙e doing at MagicSchool is we, we have an enterprise version of Claude that our employees use and where we have a real big focus on like agent decoding for our engineers to move faster and ship with more velocity, build a lot of really amazing things for our users. And one way we鈥檙e measuring the efficacy of generative AI right now is we鈥檙e looking at our leaderboard, like, which engineers are using the most tokens in our version of that enterprise dashboard? And then we鈥檙e asking the managers, would you call that your highest performer? Like, is that engineer who also is using the most tokens in Claude performing better? Are they shipping more? Are they meeting the goals that you鈥檝e set for your team in a better way because they鈥檙e using AI more? And if the answer is no, then we need to rethink about is generative AI actually helpful? But if the answer is yes, then we need to spotlight that engineer and we need to have that engineer teach the other engineers how they鈥檙e using it and how it鈥檚 making them more productive. And Aurora was doing the exact same thing two years ago. So I think it鈥檚 a really, really cool way to think about how this is actually amplifying productivity in a meaningful way.

Michael Horn: Adeel, I think that鈥檚 a good place to leave the conversation for now. I appreciate how much you鈥檝e geeked out with us, and also I appreciate that you鈥檝e told us where it is now and where your sort of vision for it is. As you know, a lot of entrepreneurs, they sell the future/present as one package as opposed to distinguishing the two. So I appreciate that in this conversation as well. Before we move to our last segment, as listeners know鈥�

And with that we鈥檒l move to our last conversation, which is it鈥檚 just a fun segment we鈥檝e had and people track this and so forth, Adeel on LinkedIn and believe it or not, about things that we鈥檝e been watching, listening or viewing and would love to hear something that you鈥檝e been tuning into what鈥檚 either on your bedside table or on your playlist.

Adeel Khan: I think I鈥檒l go with a Netflix show. So it took me quite a long time to get to it, but I finally watched just finished the last season of Stranger Things, which I was late to the party yard in the first place but kind of binged it a few years ago. And so I was eagerly anticipating the final season. None of it got spoiled for me and I got to watch the whole thing and it was delightful. So I felt like I wrapped the bow on a really special I feel Stranger Things is so awesome if you guys have seen it, but I feel like it鈥檚 such a special cultural phenomenon that everyone kind of experienced together and watched together. So that was mine. Speaking of a new show. So excited to hear from you guys.

Diane Tavenner: Well, I鈥檓 not going to give you a show today. I apologize. I鈥檓 gonna give you a book. It鈥檚 a little bit ironic to have AI books, I think, but this one feels special to me. So it鈥檚 called Co-Intelligence Living and Working with AI by Ethan Mollick. And I will say, when it arrived, my kiddo who works on the models said, oh, that鈥檚 perfect for people like you. I was like, what do you mean? He鈥檚 like, you know, for, for regular people who don鈥檛 understand what we鈥檙e actually doing, but, like, have some sense of it. And it鈥檚 really useful for kind of who are really trying to make sense of AI and world and what that looks like.

And that鈥檚 what it feels like to me. And so it鈥檚, you know, I鈥檓 not telling you anything new by sharing this book with you, you know, a bestseller. But it is, it鈥檚 short and it鈥檚 thoughtful and I think useful for anyone who鈥檚 really trying to grapple in this space. I would recommend it.

Michael Horn: The one question Diane, I have, I love Ethan鈥檚 Substack as well. And when the book came out, I bought it as well and read it. But I鈥檓 just curious, like, does it still feel current given, like, the race?

Diane Tavenner: You know… yes?

Michael Horn: Interesting.

Diane Tavenner: Yeah, I think so.

Michael Horn: OK.

Diane Tavenner: I think so.

Michael Horn: Cool.

All right. I鈥檝e got a book as well. So, Adeel, I鈥檓 also striking out on the show, watching, I think, at the moment, but I still haven鈥檛 done Stranger Things. Diane, I think this was on her list. I can鈥檛 remember how many episodes ago

Diane Tavenner: I started early, but I haven鈥檛 finished off the season. So you鈥檝e been 鈥�

Michael Horn: I was about to say. You just inspired her to finish it. Yeah. The book I鈥檝e been reading is A Heart of a Stranger by Angela Buchdahl. She鈥檚 a rabbi at the Central Synagogue in New York City, which I think is the largest synagogue in the U.S. or, or top two, I guess. And it鈥檚 terrific, she鈥檚 a Korean American rabbi. And so it鈥檚 like a very interesting story about where she, the circle she has not belonged in, and then making sort of this, I guess, momentous movement into being a rabbi and sort of what that鈥檚 been like and her life story through it.

So it鈥檚 been a very good read. As Diane knows, my wife鈥檚 Korean American, so, like, and I鈥檓 Jewish, so it鈥檚 like hitting on multiple levels in our household. TBD if anyone else in the household reads it beside me. But I鈥檓 enjoying it quite a bit. And we鈥檒l leave it there. Adeel, huge thanks for coming on, joining us, geeking out with us, and for all of you, keep the questions, comments, notes coming after this episode and in general, and we鈥檒l see you all next time on Class Disrupted.

This episode is sponsored by LearnerStudio.

The middle schoolers at Philly鈥檚 Marian Anderson Neighborhood Academy have a lot of questions about artificial intelligence.

They want to know how the government is using AI and what impact the technology has on the environment. They鈥檙e curious about how it鈥檚 being used for creativity, and whether it will be with us forever 鈥� or if it鈥檚 an economic bubble waiting to burst.

The sixth through eighth graders have been researching these topics and grappling with how it makes them feel about themselves, their education, and the world around them. On Friday, they presented their findings to their parents, teachers, and some state and local officials in their school cafeteria. Overall, they said there鈥檚 a lot they still don鈥檛 know.

Sixth grader Azizah Simmons said she鈥檚 weighed the pros and cons and she鈥檚 pretty confident that AI鈥檚 overall effect on our society is negative. If used correctly, Simmons said language models like ChatGPT could help kids her age improve their writing. More often than not, she said students use it to cheat on homework or cut corners on writing assignments.

But it鈥檚 the ubiquity of the technology that worries her most.

鈥淵ou can鈥檛 really escape AI,鈥� Simmons said.

Conversations about AI have permeated every aspect of education since the arrival of models like ChatGPT. Familiar debates about cheating have given way to Marketing pitches from companies promising 鈥渢ransformative鈥� AI tools are now . In Philly, educators are working with students to build their own curriculum to and that can be embedded deep in the internal code.

And students say they feel like they have as much knowledge 鈥� or sometimes more 鈥� than the adults in their lives.

Sixth graders Thomas Mapp and Tyshaan Anderson鈥檚 research project focused on how video game designers use AI for level design, character creation, and visuals. Outside of school, they鈥檝e been using AI to help them code games in Roblox and edit videos.

Anderson said he thinks the technology has helped kids like him experiment with creative fields like game design without needing to know the ins and outs of specific coding languages.

Marian Anderson Principal Nicole Patterson said she鈥檚 been inspired by her students鈥� civic inquiries and has learned a lot from them.

Patterson said she sees her school as a trailblazer in leading challenging conversations about AI. But she cautioned that 鈥渢his is unfinished work.鈥� She said students will continue their research and keep talking about these issues.

Marian Anderson computer science and technology teacher Trey Smith said the goal of Friday鈥檚 event was to help students and parents discuss how AI is now part of society, culture, politics, and everyday life, not just about how AI works.

鈥淲e鈥檙e all still trying to figure this out together,鈥� Smith said. 鈥淔or students to be in dialogue, not just with themselves and each other and me, but also with their families and with legislators and with school district officials and professors 鈥� I think it鈥檚 so important for them to learn together.鈥�

That learning process can be tricky. Simmons said she ends up using AI involuntarily because search engines like Google now frontload AI overviews. That makes it difficult for young users to differentiate between what is a primary source link and what is AI generated.

鈥淵ou use it without meaning to. It鈥檚 everywhere implanted in our lives,鈥� Simmons said.

Chalkbeat is a nonprofit news site covering educational change in public schools. This story was originally published by Chalkbeat. Sign up for their newsletters at .

Yet just freely letting students use AI to generate ideas, explain difficult concepts and produce/revise writing 鈥� upon which learning depends and . 

So I have been attempting the 鈥渢hird option鈥� recommended by both the and the : teaching AI literacy.

This year my 10^th and 11^th grade English students used AI itself as a text to advance their critical thinking skills. We still read novels and short stories, still engaged in discussions and wrote essays, but AI was now a regular part of our work together.

As we read, we examined how large language models鈥� recycled novel 鈥渁nalyses鈥� mis-read and oversimplified complex literature, producing distillations that often lacked nuance compared with the creative, discursive yet defensible readings that the students themselves generated. They learned to discern actual analysis from simplistic summaries, and to suspect the allure of AI鈥檚 instant 鈥渃orrect answers.鈥�

As we engaged in literary discussions, we sometimes invited chatbots into the conversation; many students described these interactions as 鈥渂izarre鈥� and 鈥渄isjointed,鈥� adequate for reviewing plot but too circular or directionless for genuinely provocative dialogue. ChatGPT鈥檚 sycophancy in particular tended to kill the necessary tension for true debate. One student 鈥渟tarted purposely saying dumb things just to see how GPT would still find a way to say `great idea.鈥� It just felt so fake.鈥�

As we wrote, we compared LLM-generated essays with human-generated ones, teasing out how AI鈥檚 鈥渟ophisticated-sounding鈥� yet 鈥済eneric鈥� prose differed from the 鈥渕essier鈥� but ultimately, in the students鈥� judgment, more engaging language they themselves created. In a world where everyone has access to LLMs, these students were discovering the value of developing genuine voice. I hope at least some emerged thinking ChatGPT was best reserved for inter-office memos and letters to one鈥檚 utility company.

As we researched, we studied how AI search summaries 鈥� which users are now 鈥� don鈥檛 actually represent internet searches, but instead reflect word proximity within a static corpus of text, a corpus lacking access to paywalled scholarly research and therefore drawing disproportionately on unregulated chat forums. 

Students examined whether LLMs accurately reported their sources and to what extent AI drew from ideologically extreme sites. They saw how wording a query 鈥� e.g., 鈥渋s abortion safe鈥� vs. 鈥渋s abortion murder鈥� 鈥� could lead to politically-slanted results based on what the AI thought they wanted to see, and how sources often said something very different than AI summaries claimed they did. 

As we took and organized notes, students compared their manual note-taking process to the output of AI note-taking tools, learning how what we choose to include or exclude in summarizing notes, how we use emphasis and phrasing 鈥� did Africa under colonialism 鈥渇uel worldwide industrial production鈥� or were African resources and peoples 鈥渆xploited for the benefit of Western industrial profit鈥� 鈥� create and propagate different narratives.

These narratives do not ; 鈥渨hat is ranked at the top鈥� of AI searches 鈥渋s ultimately influenced by the priorities of LLMs鈥� shareholders,鈥� so we studied studying the politics of AI magnates like Sam Altman and Peter Thiel, learning how Gemini鈥檚 responses to political questions, and studying algorithmic bias (e.g., image generation requests for 鈥渄octor鈥� returning mainly white males), all helped my students re-think their understanding that AI tools were neutral and simply utilitarian.

When we studied AI, we simultaneously studied neurological research about how humans, unlike LLMs, don鈥檛 just rely on pattern recognition, but also make intuitive leaps, and used Edward De Bono鈥檚 activities as practice. Students did something else that AI couldn鈥檛: related classroom content to personal experiences. 

One multilingual student recalled attending a business meeting with her father where he faltered, because he 鈥淸knew] that someone who has the ability to speak English better [me] sat right next to him… 鈥業t makes me want to depend on you鈥� he told me, 鈥榳hen I鈥檓 totally capable of doing so by myself.鈥� He did much better after I left.鈥� The student then made the leap to consider how, even if AI help is readily available, perhaps we gain something by refusing to rely on it.

When I abandoned AI bans, I instituted AI audits. Students had to demonstrate their thoughtful, detailed evaluation of each AI tool they used, including knowledge of how it operated, what they felt they gained and lost by using it, how they verified accuracy of information, and how they had not relinquished their own thinking. The students didn鈥檛 necessarily conclude 鈥淎I is always bad,鈥� but they did see that using it always requires vigilance. Best of all, they didn鈥檛 have to take my moralizing word for any of this; they discovered it for themselves. 

Yes, I had to teach fewer novels in order to make room for AI literacy, but ultimately my job is not to teach novels; it鈥檚 to teach students. Their insights 鈥� how Grammarly鈥檚 鈥渃orrecting鈥� language altered integral parts of people鈥檚 unique voices, how personal evolution often comes from struggle and discomfort, how our desire for ease can hold us back from achieving our potential, how dangerous it is to invest authority in words just because they emerge from a machine 鈥� were equally valuable as any takeaway they gleaned from novels. And this time I knew those takeaways were theirs, not ChatGPT鈥檚.

I teach an affluent population, but are with more economically and linguistically diverse learners. To be sure, my experience was often fraught. Some of my less-confident students never stopped considering LLMs鈥� 鈥渃lear鈥� and 鈥渨ell organized鈥� writing superior to their own, and still hesitated to trust their own readings of literature over 鈥渢he answers鈥� ChatGPT offered. 

I struggle with asking students to critically evaluate AI while their own linguistic and analytic skills are still developing, but I also know I cannot create the conditions that allow teenagers to become master writers and thinkers before they are exposed to AI; they will soon arrive at my classroom having been using it since childhood. 

Post-pandemic suggests that, when teaching anything, we cannot wait for students operating well-below grade level to 鈥渃atch up鈥� before introducing higher order thinking skills; we have to figure out how to teach both simultaneously.   

That requires creativity, and creativity is what makes humans superior to AI, which can only regurgitate already-created ideas. Teachers excel at creativity; every day we come up with new ways to meet the ever-changing needs of our students, and right now AI literacy is one of those needs. 

that this training is crucial for keeping AI users 鈥� a population swiftly becoming synonymous with 鈥渉umans beings鈥� 鈥� from engaging in 鈥渃ognitive surrender, marked by passive trust and uncritical evaluation of external information,鈥� as opposed to 鈥渃ognitive offloading, which involves strategic delegation of cognition during deliberation鈥� when using AI.

about AI rendering English classes obsolete forget that the humanities are about studying what is human about us 鈥� including both our criticality and our adaptability.

Note: This is an abridged, non-scholarly version of a peer-reviewed article slated for publication in Issue 115.6 of NCTE鈥檚 .

AI assistants like Claude, ChatGPT and Gemini won鈥檛 change these realities. But when you鈥檙e overwhelmed, they can help you streamline some of the most tedious aspects of your work. They can free up your energy for what only a human teacher can do. And AI assistants can actually push us to be more creative. They can help us overcome teaching ruts, nudging us to revitalize aspects of our teaching that are growing stale.  

AI assistants have arrived at a time when teachers need support to do their best work. In a national by the RAND Corporation, just 24% of teachers reported being satisfied with their total weekly hours worked, and 66% said their base salary was inadequate.

AI tools won鈥檛 make up for unfair compensation. But they can help us save time and create a better work/life balance. They can also help us do better work. 

A 60-second guide to prompting

The first step to making the most of AI is understanding how to use prompts. A prompt is a natural language instruction to an AI assistant. It doesn鈥檛 have to include technical or formal language. You don鈥檛 even have to use full sentences. 

Prompts are tool agnostic, so you can use them with whichever AI assistant you have access to. I recommend the free or paid versions of Claude, Gemini and ChatGPT, but you can also use free AI tools that run privately on your own laptop, like or . 

As the teacher, you guide an AI assistant like Claude or Gemini with relevant context. For prompts to work well they have to be detailed, including specifics and context. Generic prompts yield generic responses. 

It helps to iterate on AI output with follow-up prompts. Ask for more specificity or detail. Adapt the response for your students, don鈥檛 use it as is. Part of retaining your agency in the process is making sure you build on whatever outputs an assistant generates. You鈥檙e the director. It鈥檚 much like you were adopting an . The advantage, though, is that this material will be more tailored to your students and teaching approach.

Knowing how to use prompts effectively can mean the difference between AI that鈥檚 actually helpful and AI that鈥檚 gimmicky. The prompts below are designed to provide you with a creative boost. They each illustrate a practical way to use AI in support of thoughtful, pedagogically-sound teaching. You don鈥檛 need any special technical skills or subscriptions. You can copy, paste, and customize them to suit your subject matter.

Setting up a project

If you want to use prompts more efficiently, create a Claude or ChatGPT Project, or a . That鈥檚 a folder where you provide a summary of context about your students that the AI assistant can reference whenever you ask for support. You can also upload past materials, syllabi, lesson plans, curriculum guidelines, Common Core State Standards, or whatever else would be helpful context for the AI assistant. 

You can also provide detailed instructions in the project for how you鈥檇 like the AI to assist you. You鈥檙e training the AI assistant and teaching it your preferences. Once you set up a project,  you won鈥檛 have to repeatedly type in the same context. I set up projects for each of the classes and workshops I teach.

The Prompt Collection

The Bell Ringer 

Start class with a spark

The first five minutes of class set the tone for everything that follows. A short, well-designed opening activity can draw students in. Engaging openers are especially valuable on Mondays, after vacations or when you鈥檙e pivoting to a new topic. The challenge is coming up with fresh ones regularly. 

Goal: Generate a bunch of quick activities you can use at the start of a class session, adapted for your subject and your students. 

Prompts: 鈥淚 teach [subject x] to students in [grade level x]. We鈥檙e studying [specific topic xyz. Be as detailed as possible about your subject and context. Include a sentence or series of phrases of context about your particular class and teaching style, or any special needs or context for your students. No need to make it formal].鈥�

鈥淕enerate five bell-ringer activities I can adapt to open a [xx] minute class. Each should take no more than [x] minutes, require no materials, and either activate prior knowledge or help students reflect on what they鈥檝e just learned. Include one that鈥檚 discussion-based, one that鈥檚 written and one that鈥檚 a game or visual/creative task. [Or adapt these examples to reflect your subject matter. For example, one of the options could be a logic puzzle or an artistic challenge].鈥�

Prompt Example

The Real-World Hook 

Answer 鈥淲hy does this matter?鈥� before students even ask

When you鈥檙e juggling administrative meetings, multiple preps and paperwork, it can be hard to give extra attention to helping students relate to a given learning unit. This prompt helps you brainstorm connections to contemporary music, art, film, TV, cultural trends or other subjects of interest to students. 

Goal: To generate five ways to show your students how the topic you鈥檙e teaching is relevant to their lives, each with a two-sentence hook you can use to open discussion.

Prompt: 鈥淚鈥檓 about to begin a unit on [x topic] with [x grade level] students. [Provide a sentence of additional context and a few additional details about your students鈥� interests]. Generate five ways to connect this material to something students at [x] grade level may likely be able to relate to. This can include sports, the arts, social media trends, pop culture, music or other contemporary issues. For each connection, suggest a two-sentence hook I can adapt to help jumpstart a class discussion.鈥�

Prompt Example

The Bad Example Generator 

Turn common mistakes into teachable moments

Showing students examples of common mistakes can help them avoid those pitfalls. But we can鈥檛 embarrass students by showing examples of their weakest work. Fortunately, AI assistants are excellent example generators. They can come up with nearly any kind of error you specify, saving you hours you might otherwise have spent creating intentionally bad work.

You can adapt this prompt to include any kind of error you want your students to avoid. These can include experimental design mishaps in science or mangled math formulas. If you鈥檙e teaching essay writing, showcase logical fallacies or ad hominem arguments. 

Here鈥檚 an example of a I generated with the help of an AI assistant.

Goal: Produce five realistic examples of a specific error type, unlabeled, so students can identify, discuss and learn from the flaws. 

Prompt: 鈥淚鈥檓 teaching [x subject/topic] to [grade level x] students. [Provide an additional sentence of specific context about your class, the learning goals you鈥檙e focusing on, and/or the lesson you鈥檙e preparing.] Generate five examples of paragraphs with [ad hominem arguments / circular reasoning / weak thesis statements / misleading use of statistics / or pick any other weakness] related to [x topic]. Make sure each example is realistic and plausible. These should be the kinds of errors students at this grade level might actually make. Don鈥檛 label what鈥檚 wrong. I鈥檒l use these for a class activity where students identify and explain the flaws themselves. [You can also task the AI with annotating or explaining these errors to help you walk students methodically through these common flaws.]鈥�

Prompt Example

The Scaffolding Prompt 

Make instructions clear for every student

Complex instructions often trip up students. Simplifying language can help, along with breaking guidance into smaller steps. This prompt helps you clarify instructions for an existing assignment, handout or any other activity. It鈥檚 particularly useful if you have students with learning differences or if your class has a wide range of readiness levels.

Goal: Reframe an existing handout or assignment so it鈥檚 clearer and more accessible, especially for students who need extra support. 

Prompt: 鈥淗ere is a [handout / assignment / resource] I give students: [paste or upload the handout]. Help me reframe this for students who face [specific challenges or context that impact some of your students]. I particularly want this to be more accessible for students who need extra support. Break the instructions into smaller, numbered steps. Replace any abstract language with concrete, specific directions. Point out any parts I should clarify. Suggest a brief example for each major step and any illustrations or images that might help me make this more visually engaging. Maintain the academic expectations I have for the work. The goal is clarity, not simplification.鈥�

Prompt Example

The Review Game Generator 

Create engaging questions efficiently for learning games

Coming up with a long list of review questions can take hours and designing multiple plausible wrong answers for every question can be exhausting. An AI assistant can help, quickly turning existing handouts, lesson plans or fact sheets into engaging questions. It can help you customize questions for your subject matter and student level. 

Goal: Generative 15 multiple-choice review questions, tiered by difficulty, formatted for whatever learning game you prefer. 

Prompt: 鈥淚鈥檓 finishing a unit on [x topic] with my [grade level x] students. I鈥檓 preparing an end of term review session, so I鈥檓 trying to come up with some good questions to help students practice [a particular skill or area of knowledge].  Generate 15 trivia questions based on the following key concepts: [list concepts or paste notes or upload a handout]. Suggest a series of multiple-choice questions, each with a correct answer and three plausible wrong answers. Vary the difficulty鈥攆ive easy, five medium, five challenging. Flag the correct answer for each. Also suggest some true/false, fill-in-the-blank, and open-ended questions for variety.鈥�

Prompt Example

The Fresh Angle Search 

Bring new life to familiar content

Some topics get stale, especially when you鈥檝e taught them the same way for years. To liven up an old lesson, it can be helpful to gather new sources, examples, statistics, case studies or unexpected angles. 

Use , a free, AI-powered search engine that provides citations alongside its results. The links it provides ensure you have an evidence trail you can use to verify its responses and to dive deeper. Digging into Perplexity鈥檚 concise search summary is more efficient than sorting through hundreds of blue Google links.

Goal: Find five recent or unexpected real-world examples of a concept you鈥檙e teaching,  including perspectives from outside the U.S. and connections to students鈥� current interests. 

Prompt: 鈥淚 teach [x topic] to [grade level x] students. [Provide additional context here about the topic or learning outcomes you鈥檙e focused on]. I鈥檓 looking for interesting material [or whatever other description you prefer] to make this subject more engaging for students. [Include any additional context about your students鈥� interests]. Find me five recent, unexpected or counterintuitive real-world examples of [x concept] that might surprise or intrigue students. Include also several real-world details to help add nuance for students who think they already understand the concept. And suggest several new analogies I can use for students who don鈥檛 yet understand this concept. Include international examples, and at least one that has an element of humor.鈥�

Prompt Example

The Skeptical Student Prompt 

Prepare for the hardest questions before class starts

You never know what odd questions might arise when you teach a new topic. AI assistants can help by generating all sorts of potential questions. That prep can help you avoid unpleasant surprises in class, so you鈥檙e ready for nearly anything students might toss at you.

Goal: Generate 10 challenging questions a skeptical student might ask me about this lesson. 

Prompt: 鈥淗ere is a [lesson plan / reading / concept] I鈥檓 teaching: [paste or upload material, mentioning the grade level and any other relevant context]. Give me a list of potential student questions about the relevance of this new topic and about real-world applications. Include also a mix of other unusual or surprising questions curious students might ask. If these high school students doubt this material is relevant, what might they ask, and what aspects in particular might they question. Generate 10 challenging questions students might ask. Include questions that challenge the relevance of the topic, the reliability of my sources and the assumptions behind my explanations.鈥�

Prompt Example

The Blind Spot Audit 

Find your own blind spots before students do

During a typical week, we don鈥檛 always have time to trade peer feedback on lesson plans or syllabi. But we can still benefit from getting input on our materials. AI assistants can critically evaluate your materials for clarity, accessibility, inclusivity or other blind spots. You always have the option of ignoring the observations. I find that many of the weaknesses the AI assistant points out are ones that benefit from a fix.   

Goal: Identify specific places in your lesson plan or syllabus where I might have an unconscious bias, where my instructions may be unclear, my examples may not reflect student diversity or my assessment criteria might be confusing. Or point out unnecessary jargon.

Prompt: 鈥淗ere is my [lesson plan / syllabus / unit overview]: [paste or upload document]. Take the perspective of a critic with expertise in inclusive pedagogy and student-centered design. Identify parts of my plan that may not work for someone with physical differences such as a vision, hearing or mobility impairment. Also point out places where an unconscious bias might be influencing the way I鈥檓 presenting this topic. Point out places where examples or explanations I鈥檝e included might not make sense to my diverse students. Show me places where my assessment criteria could be made more clear. Note any other sections of the material that might not be inclusive, accessible or relatable for students. Be direct. Include the location of each issue so I can explore potential fixes. I want specific critique, not general praise, and I want you to explain each observation in detail.鈥�

Prompt Example

The Differentiation Prompt 

Adapt one assignment for three distinct student levels without tripling your prep time

In many classrooms, students arrive at varying levels of readiness. Creating three versions of the same material is one of those things that turns a 40-hour week into a 53-hour one. Tasking an AI assistant with suggesting adaptations of your material ensures that your newly differentiated materials will remain anchored in your own ideas and teaching goals. 

Goal: Produce two alternative versions of an existing assignment: one with additional scaffolding, and one with stretch challenges for advanced students.

Prompt: 鈥淗ere is an [assignment / assessment] I give students: [paste or upload material]. Generate two versions of this: one for students who need additional scaffolding and more explicit guidance, and one that adds stretch challenges for advanced students. Preserve the core learning objectives. Summarize the suggested changes and explain their rationale, so I can decide how to adapt these alternatives for my students.鈥�

Prompt Example

The Rubric Builder 

Help students understand how you鈥檒l assess them.

A well-designed rubric does two things: it clarifies your expectations before students start working, and it gives them a roadmap for revising. Developing rubrics from scratch is tedious. It requires formatting small batches of text into boxes in complex tables. This prompt generates a structured first draft in table format. You can then refine it before sharing it with students. To start, specify the elements of the student work you鈥檒l be evaluating, and describe your criteria. 

You don鈥檛 have to use full sentences or formal language. Just describe what constitutes excellence for this assignment, what satisfactory work looks like, and what evidence signals to you that a student may need more skill practice. Developing these rubrics with AI assistance is an iterative process. Revise initial outputs by adding your own details and refinements.

Goal: Generate a rubric with three performance levels and five criteria you鈥檝e specified, written in specific, concrete language, without vague phrases like 鈥済ood use of sources.鈥�

Prompt: 鈥淚鈥檓 assigning [describe assignment] to [grade level x] students. [Provide any additional relevant context]. Generate a rubric with three performance levels: Excellent, Proficient and Developing. Include five criteria relevant to this assignment: [list criteria, e.g., argument clarity, use of evidence, originality, structure, mechanics]. For each criterion and each level, write two specific sentences describing what that performance actually looks like. Avoid vague language like 鈥榞ood use of sources.鈥� Be concrete. Put this rubric into a table, then await my input for potential edits鈥�

Prompt Example

The Case Study Collaborator 

Generate fictional scenarios to spice up discussions

Case studies help spark lively discussions. They鈥檙e useful whether you鈥檙e introducing students to ethical questions or trying to help students relate to a historical situation. They can also be useful for bringing a business decision or a scientific discovery to life.  Creating cases from scratch can be exhausting. So this prompt helps you build fictional but realistic scenarios customized to your subject matter and student context. 

Goal: Create a fictional case study to illustrate a tension relevant to your subject, set in a context students can relate to, ending with three discussion questions.

Prompt: 鈥淚 teach [x subject] to [grade level x] students. [Provide an additional sentence of context or specifics to ensure the case studies are relevant and useful.] We鈥檙e exploring [x concept or issue. Include as much detail as possible about what and how you鈥檙e approaching the topic and your learning goals]. Create a fictional but realistic case study involving [type of character, institution, or situation relevant to your subject] that illustrates the tension between [value A] and [value B]. Set it in [context relevant to your students鈥攁 school, a local community, a specific industry]. The scenario should be complex enough that reasonable people could disagree about the right response. End with three discussion questions that I can adapt to push students to apply the concepts we鈥檝e been studying.鈥�

Prompt Example

Disclosure: Two kinds of prompts appear in this piece. I developed the templates with brackets based on my teaching experience. The filled-in examples showing how teachers might customize each template were drafted with help from Claude, an AI assistant. Using AI to help generate these examples let me stress-test and customize each template across different subjects and grade levels and confirm that the prompts produce useful results. I reviewed and edited every example.

Teacher Jake Baskin remembers exactly where he was when he first watched the that introduced to the world, inviting kids to learn how to code. 

鈥淚 was sitting in my high school classroom in Chicago,鈥� he said. 鈥淚 got a link to that first video and thought, 鈥業鈥檓 so excited. Someone else is saying the things I’ve been saying to my students.鈥� 鈥�

A longtime educator who now leads the , he watched as the nearly-six-minute video showcased Bill Gates, Mark Zuckerberg, Jack Dorsey and a constellation of tech celebrities recalling their first experiences with a computer: creating games, drawings, quizzes and more. 鈥淚 was 13 when I first got access to a computer,鈥� says Gates, a wistful smile crossing his face. 

It didn鈥檛 hurt that he and a few others onscreen were by then among the wealthiest people on the planet.

The video soon helped spark what would become arguably the most successful education reform campaign of the past few decades.

By 2021, offered computer science, known widely as 鈥淐S.鈥� persuaded legislators in 12 states to add it to their high school graduation requirements. And every U.S. president since 2013 has made computer science a pillar of their education agenda.

Baskin liked the video so much he鈥檇 go on to spend four years at Code.org, helping the nonprofit write its first curricula and building district partnerships nationwide.

But fast-forward to 2026, and the landscape looks more fraught. So-called Silicon Valley 鈥溾�� have spent the past few years secretly building and while of software engineers. And the organization that made 鈥渓earn to code鈥� a national rallying cry must confront an existential question: In an era when generative AI tools can create functional code from plain-language prompts 鈥� and where kids are making millions 鈥渧ibe coding鈥� professional-looking apps 鈥� where exactly does a nonprofit called Code.org fit in?

New CEO Karim Meghji admitted that he and his colleagues must reframe their offerings and message without abandoning their core ideals. 鈥淥ur foundational principle is not, 鈥楳ore kids need to learn how to be software engineers,鈥欌�� he said in an interview. 鈥淲hat we’ve been promoting is that a world that is very digital, and has technical products all around us is a world where students deserve to understand how these things function, how they work.鈥�

That reframing comes at a key time for the nonprofit, whose gift-fueled funding has in recent years, from $42.8 million in 2023 to $25.2 million in 2025. It reflects both shifting philanthropic priorities and the existential questions now swirling around the field of computer science. 

Is computer science collapsing?

The shift Meghji describes is happening not just in K-12 education, but in the higher ed landscape and in the broader job market.Student enrollment in computer science at four-year colleges last fall, the biggest single-year drop of any major discipline since at least 2020. In one year, computer science fell from the nation鈥檚 fourth-largest undergraduate major to its sixth, even as the fortunes of Silicon Valley . 

At the University of California, computer science graduates are expected to number about 350 next year, from 2025. Across the entire UC system, computer science enrollment declined last year for the first time since the early 2000s.

The job market for young coders has softened, too. A recent study by, using payroll data from millions of workers, found that by September 2025, employment for software developers aged 22 to 25 had declined nearly 20% compared to its peak in late 2022 鈥� even as employment for more experienced developers held steady or grew. The study’s authors described entry-level engineers as 鈥渃anaries in the coal mine,鈥� early casualties of AI tools that can easily replicate their work.

Other data paint a less clear picture. A by the finance analysis firm Citadel Securities found that in the long term, software developers鈥� jobs may be relatively safe because replacing them en masse with AI would require 鈥渙rders of magnitude more compute intensity鈥� than the industry has. Alex Kotran, CEO of the , noted that job postings for software engineers are actually up 11%.

鈥淪omething that I just want to shout from the rooftops, is, 鈥榃e really don’t know what is about to happen,鈥� 鈥� he said.

That uncertainty, it turns out, is what Meghji is emphasizing as Code.org shifts direction. 

Yes, AI seems miraculous and it鈥檚 improving quickly. But it also fumbles on occasion, , and generally threatening to on the world. Meghji invoked the notion of AI鈥檚 鈥�,鈥� which describes its strange, counterintuitive competence in complex processes 鈥� but that can also fumble . 

For Meghji, a veteran consultant and technologist who most recently was Code.org鈥檚 chief product officer, that jaggedness is exactly why teaching computer science matters now: 鈥淭he further we move away from how these systems work 鈥� the further we abstract away from what’s happening under the hood 鈥� the more important it is that students learn foundational CS and computational thinking concepts,鈥� he said.

When AI shows its fallibility, he suggested, educators should view it as a teachable moment.

As it rebuilds, his organization plans to keep coding at its center while weaving AI into instruction, Meghji said. It has replaced its well-known 鈥溾�� with an Hour of AI, and it鈥檚 developing an 鈥淎I Foundations鈥� course for high school students, due this fall, in which students use AI to help build and lay out interactive websites, then use a combination of their own written code and AI-generated code to improve the sites. A middle school curriculum is also planned.

鈥淲e don’t start with AI,鈥� Meghji said. 鈥淲e start with the foundation, teach the principles. Then we introduce AI coding, have students read code that AI is generating, find the issues, and hopefully have a higher ceiling 鈥� both in terms of their creative output, their agency, and what they鈥檙e producing.鈥� He estimates that where previously perhaps five out of every 100 students built something genuinely impressive, AI tools could raise that to 30 or 40.

He鈥檚 also tweaking the organization’s business model. With philanthropic funding down sharply, Meghji said, he鈥檚 exploring whether Code.org can generate earned income through curriculum offerings tied to dual-credit and career and technical education pathways, models where public funding could help students earn technical credentials. He wants its curriculum to remain free for students but is exploring state and federal funding to underwrite it.

鈥楢 fool’s errand in any field鈥�

Meghji is also eager to correct a misconception that he believes was never really Code.org’s message: the idea that learning to code was to a six-figure salary. 

鈥淥ur message was not, 鈥楬ey, come to Code.org, take computer science, and you’re going to write your ticket,鈥欌�� he said. 鈥淲e’ve always been of the mindset that every student deserves the right to learn the foundations of how technology works.鈥�

Baskin, the former computer science teacher, said he wishes that distinction had been drawn more sharply from the beginning.听

鈥淚f I could go back in time, I would try to keep the movement from explicitly linking computer science to short-term career outcomes, because that’s a fool’s errand in any field,鈥� he said. 鈥淣o one knows what the jobs of the future will be like, and if they did, they’d be very, very rich. It’s about preparing students for the things we don’t know that are coming and giving them the broadest opportunity to engage in what is meaningful to them.鈥�

aiEDU鈥檚 Kotran made a similar case, arguing that computer science should sit 鈥渁longside reading and writing and math and science,鈥� not as vocational training but as the place where students practice so-called 鈥渄urable skills鈥� such as collaboration, design thinking, productive struggle and iteration. 

He worries about the consequences if schools abandon the field entirely. 鈥淚f we turn our backs to computer science, you’re going to have this deviation where kids who have access to those learning experiences are just going to be on a separate track,鈥� he said, with access to knowledge that others don鈥檛 have. That鈥檒l worsen inequality.

The strongest case an organization like Code.org can make, Kotran said, is actually a counterintuitive one: That AI, the very technology threatening to upend coding careers, might actually help recruit the next generation of computer scientists.

Despite the appealing creation myths embedded in Code.org鈥檚 famous intro video, he said most young people who study computer science must put in upwards of two years before they get to a place 鈥渨here you could build something that’s actually cool.鈥� But many students never made it that far. With AI, the time horizon shrinks: 鈥淵our first class is like, 鈥極K, let’s vibe-code something. Think of a problem you want to solve that’s relevant to you 鈥� finding the right makeup, predicting fashion trends, sports data analytics, whatever,鈥欌�� he said. 

Students build something, but to further develop it, they need to go deeper and understand the code behind the vibe. Code.org and groups like it could open that experience up to students for the first time. 鈥淚 don’t think we ever had something that powerful before,鈥� he said. 鈥淎nd if we wield it right, we can actually start to reach kids who don’t think of themselves as CS kids.鈥�

Updated: This story has been updated to reflect the most recently released funding figures for Code.org.

In this episode of Class Disrupted, hosts Michael Horn and Diane Tavenner turn from schools powered by artificial intelligence to the tools themselves.

Matt Pasternack, founder and CEO of , shares the company鈥檚 journey from its low-tech beginnings to an AI-powered platform for early reading instruction. Pasternack shares how Once now delivers effective, personalized one-on-one tutoring through software, while emphasizing the importance of human connection in the learning process.

Listen to the episode below. A full transcript follows.

Diane Tavenner: Hey Michael.

Michael Horn: Hey, Diane. Good to see you after a few episodes, diving deep into school models and thinking all about AI and what it enables today.

Diane Tavenner: Indeed, we are going to shift gears a bit today, not away from AI because based on all the emails, the calls, the feedback we鈥檙e getting, this is the thing folks are thinking about and talking about. And so we鈥檙e sticking with AI. Rather, we鈥檙e going to shift away from AI school models, full school models, and infrastructure to how AI is being used directly by and with teachers and students and in classrooms. And so I think this is going to be a really interesting other dimension of what鈥檚 happening.

Michael Horn: Yeah, I think that鈥檚 right. I mean, we thought this was the next logical place to go. Everyone says if you鈥檙e not changing the classroom at some point, you鈥檙e not changing much. And so we wanted to go into that classroom and explore how a small number of folks, entrepreneurs, are thinking about really how do we use AI in classrooms without the entire school itself or the system around it being changed. And so this is it. I鈥檓 really excited for this conversation. Someone we鈥檝e both known for a long time and get to go deeper on it Diane.

Diane Tavenner: A very long time. I鈥檓 really excited today to welcome Matt Pasternack to the conversation. I was trying to think about when we met Matt, but it feels like so long ago. I can鈥檛 even remember at this point, because you, you actually were early at School of One, which is now Teach to One as a director of assessment. And then you went and you were on the founding team of Clever, which many people will know as a tool that effectively connected edtech products to student information systems. So this really critical infrastructure piece that enabled so much of what we now sort of take for granted in terms of technology. And then most recently, you鈥檙e the founder and CEO of Once, which is a company that leverages, well, I鈥檓 going to say what I think it is, and then we鈥檙e going to get into it. You鈥檙e going to really describe it for us.

A company that leverages the research around tutoring by using reading-based software and human people to teach 3 to 7-year-olds to read. And so we鈥檙e just grateful to have you here. Thanks for joining us.

Matt Pasternack: I鈥檓 thrilled to be here. I鈥檝e known you both for quite some time, and it鈥檚 really an honor to be on the show.

Michael Horn: Well, we鈥檙e thrilled you鈥檙e making the time for us. So let鈥檚 dig into it. As Diane mentioned, you are working on solving a problem that we鈥檝e talked about on this podcast historically, which is teaching reading, something that there鈥檚 a lot of evidence around how to do. But as Diane said, before tackling this reading challenge, you helped build Clever. And it does feel like a big shift from, if you will, school infrastructure to teaching, learning, curriculum. Maybe it鈥檚 back to your roots in some sense from School of One. But tell us about the origin story of Once and your own personal why for building it.

Rethinking Early Reading Education

Matt Pasternack: Sure, sure. Well, I鈥檒l share two things. You know, one is that, you know, very early in my edtech career, before Clever, I worked on some projects that were very, very expansive in what they tried to accomplish, and both on the data infrastructure side as well as on the curriculum side. And it was sort of the heady days of the late aughts or early 2000s, and then the early 2010s, excuse me. And sort of this belief, if you just sprinkled a little bit of technology or software magic on things, education, everything would just work. And there were these really exciting analogies to Netflix鈥檚 personalization. I鈥檓 sure you remember those days. And so I spent some time trying to boil the ocean.

And then after that, I just kind of made an abrupt turn and never looked back. And I said, look, I want to work on specific problems in education. And the first one was what Clever tackled. It was an area I had a lot of insight into from some of those earlier explorations on, you know, the specific problems that rostering and single sign-on presented to schools, and particularly schools trying to adopt software in varying ways. After Clever, I kind of went in a little bit of a different direction for a while, focused on voting. And then in the sort of early to mid-pandemic, I was catching up with an old friend of mine who had been a teacher in the same school where I taught after college, and we were talking about how during the early pandemic, both of our kids who are kindergarten age were out of school because schools shut down. And yet both of them learned to read in that year out of school.

Now, my wife is a former kindergarten teacher, and so I was like, oh, I had an unfair advantage, you know? I mean, she had sort of one-on-one tutoring from a kindergarten teacher. Missing kindergarten, but, you know, my close friend was a former middle school teacher, and, you know, it鈥檚 not like he had been working on that expertise forever, but he said, yeah, you know, structured 15-minute lessons every day, you know, his child learned to read, and we just sort of stepped back from it and said, wait, you know, for as long as anyone can remember, we鈥檝e been teaching kids to read, attempting to teach kids to read, in these sort of 30-person classrooms, and, you know the numbers, and in sort of any education environment, about half of kids, if given some of the right foundation, will learn to read to some extent. And so any teacher can look at their progress and say, well, some of my kids are learning to read, it鈥檚 working. But if you want all of your kids to learn to read, you know, you can鈥檛 say that what鈥檚 happening today is working, because we鈥檝e had flat NAEP scores for decades. And so we just started to kind of think really big picture and say, you know, what would happen if every kindergartner got the education that our two kids had gotten in the early pandemic, which is just 15 minutes a day of one-on-one reading instruction. And initially, we actually wanted to avoid schools, avoid K-12. We said, look, the sales process in the K-12 is so complicated. Let鈥檚 do something different.

Let鈥檚 actually focus on, because preschool, you know, research was just coming out about how early the brain starts developing. For where children begin, you know, learning, you know, are able to learn to interpret written language. And then as we start to explore the preschool route, we realized that that was probably even harder than K-12. And so we kind of returned to our roots and started in a charter school, actually was our first implementation, and basically just said, hey, can we put a couple people in the back of this TK classroom and teach, you know, have them provide daily one-on-one tutoring to the students to help them learn how to read. And we had some amazing success stories, and, you know, it was just a couple-month pilot. There was a child who, a child who came to school every day, you know, lots of home context, crying and kind of hiding under his desk and did not want to be in school, And it still makes me kind of emotional to tell the story, but, you know, even a month or two in, he was sort of opening up to school and responding to school, and people asked him, you know, what had changed, and he said it was the one-on-one tutoring he was getting every single day. You know, and that鈥檚 a 5-year-old who, you know, that鈥檚 going to determine the next, you know, the rest of his life, essentially. And so, you know, we kind of took that and said, and the folks who were providing the tutoring, had not had background as reading tutors.

That was really important to us, because we said there just aren鈥檛 enough kinds of trained reading specialists in the country to do this at a scale of 4 million kids a year. So you need a method that allows any adult to access material and provide this instruction. And then we had this really interesting call with Portland Public Schools, and this is kind of right after that first, right as that pilot was winding down. And we were talking to them, and they said, well, you know, we can鈥檛, you know, we can鈥檛 afford to, you know, sort of put all these other people, you know, pay you to bring all these other people into schools and teach kids how to read. We don鈥檛 even know where you鈥檇 source them. But they said, look, we have tons of instructional assistants. That was a position that they had at the time who were supposed to be working on reading in the early grades. You know, could we use this type of program with them? And that was kind of our lightbulb moment because we said, look, you know, we are, we鈥檙e not an HR company, right? We have no expertise in how do you hire, you know, tens of thousands of tutors all over the country to provide this instruction.

We, you know, attempted it in some pilots and literally were unable to source even like a couple tutors in some metropolitan areas. And so we said, We want to work with the staff the school already has. And that really, that, you know, that conversation launched everything. We did not win that deal. We did not end up serving Portland Public Schools. It was okay. That learning that they gave us in that phone call was, you know, worth its weight in gold. And I鈥檒l always be grateful to them for that experience and taking the time with us at that early stage in our journey.

Diane Tavenner: Matt, I love that we鈥檙e going to get sprinkled into this conversation, this bonus of just what it鈥檚 like to try to build something and sell the schools and whatnot. So that鈥檚 really fun to hear that piece. Listening to your origin story, people might be saying, wait a minute, you鈥檙e spending this season talking about AI and education, but it doesn鈥檛 seem like Once is related to technology even, or AI. And in fact, you began in that pandemic period. So before the sort of famous release of ChatGPT in November 2022. And so, you know, as a guy who at face value seems to have been tech-forward, what was your sort of original hypothesis and approach to making sure, you know, to just sort of going what it seems like a full human approach to reading? And is there technology in here anywhere?

Matt Pasternack: Yeah, so it鈥檚 a great question. I think it has, there鈥檚 a two-part answer. There. The first is that one thing we were certain of from day one was that young children learn best from adults, like actual in-person human-to-human instruction. You know, for millions of years, you know, our species and the predecessors of our species have been, you know, teaching children to identify different berries and, you know, what animal is going to hurt them and which ones are safe and, you know, how to survive. And that was all done, millions of years of evolution. For, you know, in-person communication between an adult and a child, where the child was highly motivated to learn because the stakes were life and death. And sort of the idea that you would deviate away from that just sort of seems somewhat crazy on its face to us.

Matt Pasternack: So that was the motivation for, you know, you need to build trusting relationships with adults, and, you know, technology is scalable in a way that people sometimes aren鈥檛. So how can we apply technology? But let鈥檚 do it in a way that deepens that connection and doesn鈥檛 attempt to replace the connection or build a kind of robot teacher. You know, the other piece of this was that I鈥檝e had the privilege in my career, I鈥檓 not a software engineer, but I鈥檝e had the privilege to work with, you know, some very, very, very incredible software engineers. And what I鈥檝e often noticed is that when you start working with some of these folks, the first thing they鈥檒l say is, hey, let鈥檚 build a version of this product in Google Slides or Google Docs or Google Sheets. Let鈥檚 build something that鈥檚 throwaway, but that we can learn from because we鈥檙e going to waste too much time. This is a little bit before the days of live coding, but we鈥檒l waste too much time coding something up. Let鈥檚 be really, really lean in terms of how we want to model what we want to accomplish. And so our perspective at the beginning was, That鈥檚 what we want to do.

We literally started in, you know, Google Slides and Google Sheets were essentially our technology. And, you know, funders who were interested in tech staff basically ignored us because we weren鈥檛 very technical. But the one thing we said we were going to do from day one is we wanted to record every single instructional session that was given. So even though we鈥檙e talking about recordings on Google Meet, we sort of knew where things were going. I mean, yes, you hadn鈥檛 had these massive releases of OpenAI and various things, but AI was in the air, the late teens, it was in the air. Folks knew where this was going. Technology was going to fundamentally change. And we just had this belief that if you had hundreds of thousands, millions of hours of recordings of adults teaching kids to read, even if you had no other technology, that you could then, one day run software over that archive and begin to do things that others would say, oh my gosh, I wish I had this sort of database, but I don鈥檛.

So that was, I think those were kind of the two key decisions and motivations for that early, less technical beginning.

Michael Horn: Well, so let me actually jump in there then, Matt. Let鈥檚 fast forward us to where you are now. So you鈥檙e not going to be an HR solution. You say we鈥檙e not in that business. You have all these recordings. What does Once look like today? What does it do? How鈥檚 it maybe similar, different from, you know, those initial, put aside the very early pilots, but, you know, once you sort of got on the ground and running, tell us how it鈥檚 evolved from that?

Matt Pasternack: Yeah. So what we do today is we serve sort of two different audiences. We serve school districts and we actually serve parents at home. And so I can kind of talk about those a little bit in turn. But what we do, I鈥檒l start with school districts. In school districts, we solve two interlinked problems. One problem that we solve is that there are too many kids in school districts who have not effectively learned how to read. And sometimes I think the best evidence of that is when you have, you know, middle schools or high schools saying, hey, we figured out what we have to do differently.

Phonics and Staff Upskilling Initiative

Matt Pasternack: We have to start really emphasizing phonics in 7th and 8th grade or high school. It鈥檚 like, on the one hand, great. Kids who haven鈥檛 learned that do need to learn it, but that鈥檚 the evidence of an enormous problem, because there鈥檚 really kind of 1 to 2 years of phonics to learn, and if you鈥檝e had 9 years of schooling and that鈥檚 like, that鈥檚 the set of activities, that indicates just a huge, a huge, you know, a huge mess basically in the early grades. So one problem is reading, and the other problem is a lack of career ladder for entry-level staff in schools, which is also a really important problem because schools are having tremendous challenges finding a large enough teaching staff to teach the children in school. And so we said, let鈥檚 solve both those problems at once. By providing coaching and curriculum to elementary school support staff, you know, paraprofessionals, teaching assistants, instructional assistants, you know, there鈥檚 a number of different roles, as long as they鈥檙e not lead classroom teacher, because then they have to oversee 20 or 30 kids. By providing curriculum and coaching to those folks, we can upskill them to the point where anyone can provide one-on-one tutoring daily for 15 minutes to, you know, kindergartners is where we focus. We do some TK and first grade, but kindergarten鈥檚 really our focus, provide 15 minutes of daily instruction to those children.

And what this looks like, you know, kind of from a technology perspective, 鈥榗ause again, we started very low-tech, we鈥檙e now, we are a software solution today. And what this looks like is the child and the paraprofessional sit down side by side in front of an open laptop computer, We call it, we teach the parents how to build a one-desk classroom. So they have their desk, it鈥檚 all set up with the physical materials they need and their laptop computer, they sit right there. A ton of work has gone into that organization. And on the screen is, on one part of the screen is what the child鈥檚 looking at, on the other part of the screen is the script for the adult. And the adult is basically going through the script and the child is responding to the script and looking at their part of the screen. And going through a number of different science of reading-based exercises and tasks that teach the child how to read. And the adult is using a fair amount of judgment in there as well, you know, identifying when are children saying things the wrong way, when are they saying things the right way, how do you keep children motivated through this process, but the kind of focus is what鈥檚 happening on the screen.

And then we record those sessions and we provide coaching. We have a national team of coaches who then watch those recordings, and provide feedback to the paraprofessionals about how to improve, you know, essentially how to improve their instruction. Like you鈥檙e basically watching game tape of yourself and using that to improve your instructional techniques. So that鈥檚 sort of a big piece of what we do.

Diane Tavenner: Super interesting, Matt. Can you, so you鈥檙e starting in this very low-tech way with these Google spreadsheets and slides and, you know, basically prototyping what you鈥檙e gonna do. And If I remember correctly, you were videoing and then literally like doing training calls with those folks, you know, like, oh, we watch your video and let鈥檚 coach you up on this essentially. And now that鈥檚, that鈥檚 a much more seamless software experience as you just described. Take us to that moment where AI becomes the reality and you鈥檝e made this really smart bet and you have all these recordings. How does AI figure in here? And we鈥檙e trying to be really thoughtful. We鈥檝e recognized that people use the word AI or whatever that is, acronym AI, to describe almost everything, like this massive range. So the specificity is super helpful of like how you actually use it here and what role it is playing.

Matt Pasternack: So we use it in a couple of ways. All of our AI is behind the scenes. there鈥檚, So you know, this is not a chatbot. This is not, you know, either the paraprofessional asking a quick question, wait, how do I do this? Or obviously the student doing that because, you know, they couldn鈥檛 use a chatbot, they don鈥檛 know how to type yet. It鈥檚 all behind the scenes. One key way that it鈥檚 used is we are able to administer oral reading fluency tests very, very frequently on the students because we鈥檙e capturing the full session. And so then we you know, we have, have all the data, the speech data, the transcript data from that session, And so we can go and we can evaluate oral reading fluency. You know, in a typical classroom, elementary school classroom, if they use some, you know, assessment like DIBELS, the teacher might do 3 oral reading fluency assessments during the year, once the beginning, the middle, and the end.

Each one takes roughly a month because the teacher has to pull each kid aside one-on-one and administer this test. And so it鈥檚 like 3 months, you know, of instruction to some extent are spent not providing instruction, but performing these tests. And that鈥檚 just, you know, that鈥檚 not the best use of teachers鈥� time. And so simply by just having all the kids individually learning on camera, we鈥檙e able to run these tests in the background, and we have you know, results, every 2 weeks. So that鈥檚 a really important piece of AI. Again, the parents may have no idea that AI is even involved in that, and that鈥檚 great. Like, this is not you know, sort of that Gemini sparkle in the corner. This is just like, no, it鈥檚 actually just making the experience seamless.

Connected Phonation and AI Learning

Matt Pasternack: The next thing that we do is we鈥檙e huge fans of what鈥檚 called connected phonation. Again, if I鈥檓 going too deep into Tales of Literacy, let me know. But you basically, know, a lot in, in some systems, even sort of science of reading evidence-based systems, there鈥檚 a lot of focus on kind of tapping out individual sounds. So if I want to say the word bat, it would be the /b/ sound, the /忙/ sound, the /t/ sound. And when you teach a child like, oh, you just sort of say /b/, /忙/, /t/. And it鈥檚 like, well, that鈥檚 bat. Like, that doesn鈥檛 sound like bat. That sounds like /忙/t/.

You know, it sounds like these sort of disconnected sounds. And you can, you know, that is again one way that people teach it. But I think the more modern approach that, you know, University of Florida and others have really emphasized is connected phonation. So you鈥檙e not, if a B and an A come sequentially, you鈥檙e gonna teach a child to go, ba, like that B is kind of a quick sound and then they鈥檙e gonna go right into that long A sound and they鈥檙e always gonna do that when they see those sounds in sequence. They connect those sounds. Well, not only do we have audio of what鈥檚 happening in the lesson, but we also have little sliders underneath the words, and so a child can move the slider while they鈥檙e saying the word, And we can use AI to really notice things about, you know, is the child鈥檚 finger actually tracking what they鈥檙e saying? How much are they connecting? Are they pausing in the right places? Because some kids, you know, maybe they know the word 鈥渂at,鈥� so they just want to jump in and say it really fast. Well, that鈥檚 great for that word, but you don鈥檛 then learn the fundamental skills that will help you read much more complex words. So, you know, evaluating something like how quickly does a child鈥檚 finger move on the screen is another beautiful application of AI?

Then really for us, the last one, it seems obvious, but it鈥檚 just using AI to write code. It鈥檚 just we are able to move 10 times as fast on 10x fewer resources as we would be able to otherwise before this moment in time. We wouldn鈥檛 be able to do what we do without AI assisting in development. But I鈥檒l say, sort of one level deeper is, you know, where does it go from here? Is it just these sort of fluency tests and, you know, kind of tracking fingers on the screen? You know, as we progress, you know, you can really use AI to evaluate, you know, student fluency in real time, right? How did they just pronounce that word that was just said? This is a much harder technical problem than it sounds like because while the tech world has made huge advances in interpreting adult speech, where you basically try and eliminate the errors to turn some garbled thing an adult says into something intelligible. With kids, you want the opposite. You actually want to figure out which of those sounds when they pronounce that word was incorrect. Rather than automated speech recognition, you want automated phoneme recognition. And if you ask, you know, folks who are really deep in that world, everyone will admit we don鈥檛 have good automated phoneme recognition today.

It鈥檚 just not there yet. But we have, again, the video archives that will let us build towards that. So we鈥檙e very, very excited about that. You can also begin to use AI to you connect, know, reading, many folks regard as multiple strands. So there鈥檚 comprehension, there鈥檚, you know, there鈥檚 phonics, there鈥檚 phonemic awareness, there鈥檚 these different elements of it. And so how is a child progressing on these different strands, and how does progress on one strand kind of influence what you would want to teach them next? So really, that sort of learning engineering And then the final piece, just to kind of fill out the story, is, you know, a lot of folks will tell you, I鈥檝e heard on your episodes before, that, you know, 90% of learning is motivation, you know, 10% of the technology, the curriculum, but most of it is really the motivation. You know, we have, if you want to learn, between YouTube and all the different educational apps out there, like, there鈥檚 never been a better time in history to learn, and yet we see, you know, essentially flat performance over the decades. Well, one way to motivate both learners, but even more importantly, their teachers, is to show them a highlight reel of like what鈥檚 gone really well.

Well, it is prohibitively expensive for us to, you know, manually curate a highlight reel of every session of instruction. But if a pair has taught 10 kids and at the end of that day, or 20 kids, know, you they get to see, hey, here鈥檚 a 30-second recap of like your best aha moments during the day, I am coming in tomorrow. I鈥檓 not gonna call out sick. I鈥檓 not gonna do anything else. I鈥檓 going to do it, and I鈥檓 going to kind of push those kids as far as I can take them. And so that might be the most magical application of AI at all, not even kind of automated phoneme recognition, but just literally highlight reels of the incredible instruction we鈥檙e seeing.

Diane Tavenner: That鈥檚 fascinating because we鈥檝e had several conversations about motivation here, and I personally have felt kind of unsatisfied about those conversations. So this actually seems like a really amazing use of the technology related to motivation. Let me ask you about something that comes up, I think, a lot in the reading conversations, dare I say the reading wars, which is this idea that kids need to be reading things that they鈥檙e interested in and that they care about and that they鈥檙e motivated by, you know, those sorts of things. When you鈥檙e in this early stage of teaching reading, are you personalizing what the kids are reading at all? Does that matter? You know, is technology supporting that? What鈥檚 going on there?

Matt Pasternack: Yeah, it鈥檚 a great question. We are definitely keeping an eye on the ability of AI to generate what we would call, or what many people in this sort of industry would call, decodables, which are texts that students are able to decode. A lot of times things are called decodables, that actually aren鈥檛 decodable to the student. So one thing that鈥檚 really interesting about reading is that, you know, you might think, well, if a kid, you know, if they can read 75% of the words, like, that鈥檚 probably good enough. Like, that鈥檚 better than not, you know, being able to read 10% of the words or something. But actually, comprehension often requires, like, being able to read 90 to 95% of the words. So if you want to build decodable texts, like, there is not really a margin for error. You need to make sure that kids are reading text with words that they are able to decode.

Doesn鈥檛 mean they鈥檝e memorized those words. I鈥檓 not talking about sight words, but they鈥檙e able to decode those words. And the challenge of reading English, not true of all languages, but English, is that, you know, a single letter can have many different sounds depending on what other letters are around it and kind of how it鈥檚 positioned. And so if you just tell a computer, build stories using this letter or you know, using this sequence of letters, it will often inadvertently pull in the wrong, pull in the words that the kid isn鈥檛 able to decode. Well, if it pulls in one of those, maybe that鈥檚 OK, they can figure it out from context. But if you鈥檝e got, you know, a page with 15 words and 5 of them aren鈥檛 decodable, like, this decodable is not decodable. And that鈥檚 where kids can lose motivation. So it鈥檚 deeply interlaced with this concept of motivation.

So we鈥檙e keeping an eye on it. We don鈥檛 think it鈥檚 quite there yet. At the age that we鈥檙e working at. You know, I know some people working in higher grade levels, and I don鈥檛 have the expertise there, but in the kind of 3 to 6-year-old grade level, we are very, very carefully still hand-curating stories at this point.

Michael Horn: It鈥檚 fascinating, Matt, because as you alluded to, there鈥檚 this huge conversation around can you level reading? And, you know, some of that is directed at particular publishers that tried to do it in a way that was not related to Decodables, and some of it is just a broader conversation. I鈥檓 taking some new information away from this conversation, so I appreciate that. Let me shift though, rather than go too deep in that, because one of the other things that you鈥檙e doing, I think, is pushing on how AI can start to redefine the role of the educator themselves, right? In some ways, you have AI, it seems, making sure that they stick to what鈥檚 the evidence around how to teach reading the best, right? So that we鈥檙e not getting too far ahead of ourselves or freelancing in ways that may be detrimental. Then you鈥檙e also deferring to judgment, it seems, for the educator around that motivation piece and what they鈥檙e seeing on the ground with the kid in ways that you just can鈥檛 pick up with technology today. Help us understand that shift over the time around what you think that Reading Coach ultimately is and that split between technology, the human judgment, and how that gets redefined. Maybe we make it easier, frankly, or more people can be great reading coaches in the future?

Empowering Adults to Teach Reading

Matt Pasternack: Yeah, yeah, it鈥檚 a great question. It鈥檚 a great question. You know, our vision is that any adult, and by adult, actually broadened to older teenagers, are able to teach children how to read. And I think if, you know, if you look, you know, we sometimes cite a stat that, you know, 95% of kids are taught, you know, the ABCs by their parents or guardians, right? It鈥檚 like parents and guardians, they鈥檙e trying to do the right thing, They just don鈥檛 know what to do. Like teaching a child the ABCs, for a couple kids will teach them to read, for almost all other kids, it will not teach them to read. And so, and so they have, they鈥檙e motivated, they know how important this is, they don鈥檛 know what to do. So we want any adult, doesn鈥檛 matter if English is your second language, it doesn鈥檛 matter, you know, if you鈥檝e struggled to read yourself, like we want to empower any adult to be able to teach any child. Now, you know, you do sometimes, there are conflicts where, do I want to give the adult who鈥檚 delivering the instruction more autonomy so they can kind of grow more in their career, or I want to remove a little judgment and make it a little more scripted? And I think over time you鈥檒l see dials in our system where you can kind of dial it up and down either way.

Because once you get to the point where it鈥檚 just the computer speaking, well, now you鈥檙e back to just computer education with an adult patiently sitting there. That鈥檚 not how kids learn to distinguish berries, right? Like, we鈥檙e actually, we鈥檝e gone too far. So the adult needs to be really, really involved in that teaching, but, you know, understanding their capacity and what they feel like, you know, sort of, you know, because it鈥檚 not just reading a script. It鈥檚, you know, the hardest thing in teaching reading is what do you do when a kid makes a mistake, right? Like, we have a whole sort of flowchart, a whole approach we take when a child makes a mistake decoding. And being able to implement something like that is hard. And so software can help. Our biggest innovation kind of in this regard, which is something that we鈥檝e just started to roll out recently, is no longer restricting this actual teaching to paraprofessionals or instructional assistants or existing support staff at the school, but actually broadening that circle to older high school students. That鈥檚 where I sort of hinted at teenagers.

And it is, we鈥檝e just gotten started here, so we, you know, we are learning as much as they are. But I鈥檝e seen some pictures and I鈥檝e seen some videos of a high school student sitting down next to a kindergartner, teaching them to read, with the, you know, the little kindergartner just eyes full of adoration. You know, this is not a teacher, this is someone who they deeply look up to, this, an older kid who actually cares about them. And it is so inspiring. And if we had a world where our high school students could knock out reading specialization for a large number of our kindergartners, I mean, I think it would change it.

Michael Horn: Generational impact would be huge.

Diane Tavenner: I mean, yeah, when you called me with this idea, Matt, I haven鈥檛 stopped thinking about it since because I think that this is the type of thinking that we need in education right now because this checks so many boxes. I mean, not only are we putting multi-age groups together and learning, right, but we are giving, let鈥檚 talk about the impact on the high school student here who feels a sense of worth and purpose and is actually doing an early job, gaining real experience coaching and developing person. Who knows where that could possibly lead? And then you have this brilliant idea of enabling them to be entrepreneurial. Like, imagine a neighborhood where teenagers in the summer sort of have the Once tool and they can open their own little neighborhood business teaching kids to read in the summer. Like, I love this so much for so many reasons. It鈥檚 really brilliant.

Matt Pasternack: Yeah, I appreciate that. I mean, we are, you know, again, we鈥檙e learning so much right now. But the response, you know, I mentioned I kind of joked earlier about, you know, selling to school districts and how hard that was. And I mean, it really is every entrepreneur鈥檚 challenge. And in K-12, you know, there are these magic moments when school districts don鈥檛 become hard to sell to. It鈥檚 really hard to find those products. Like, it is extremely hard. Clever was one of them.

Student-Led Reading Program Growth

Matt Pasternack: I mean, Clever just blossomed across the country, and it was so exciting to see, you know, sort of help it spread and kind of watch the spread, and it was something very special. You know, we are not operating at Clever velocity right now in terms of distribution, but we have gotten such an outpouring of response, superintendents who write in saying, you know, I get a million cold emails a day, you know, I don鈥檛 read any of them, but when I saw high school CTE students teaching kindergartners how to read, I was like, oh, that鈥檚 it. You know, that鈥檚, that鈥檚 so obviously it. And we鈥檙e hearing that response. One thing that鈥檚 so exciting, to date we鈥檝e primarily focused on kind of top 500 districts, it鈥檚 very large districts. A lot of them are urban, but you know these very big school districts. And we are getting this response from tiny districts, you know, also from some large ones, but from tiny ones, from districts that can鈥檛 really afford to have paraprofessionals, but they鈥檝e got high school CTE students and they鈥檙e looking for a great thing for them to do. And they鈥檝e got kindergartners who aren鈥檛 learning how to read. And it鈥檚 like, let鈥檚 go, you know? And that is just amazing.

And so we are flying to, you know, towns across the US right now that, you know, we鈥檝e never heard their names before. And just being embraced by the folks who are there and just going right in. It鈥檚, it鈥檚 just, it鈥檚 wonderful.

Michael Horn: That鈥檚 so cool. I mean, I think about the leadership opportunities, responsibility, judgment, just like it鈥檚 checked so many boxes. And then again, the generational impact, if you do that at scale, could be humongous. Let鈥檚 wrap up with this last question, which is before we go to our what you鈥檝e been reading and watching outside of work thing. But, you know, look, Once, as you just talked about, is designed to be facilitated by, could be an instructional aide, parapro, high school student, whatever, it鈥檚 not a whole class curriculum, right? And so you talked about those magic moments where districts actually start to absorb it and so forth, but take us into the classroom itself. How are schools putting Once into their schedules and days? How are they integrating it? What鈥檚 the impact it鈥檚 having about how they think about, you know, the whole class activities that they perhaps have been doing? Are there trade-offs that they鈥檙e having to make? Help us understand where does it lock into the current schedule?

Matt Pasternack: It鈥檚 a great question. It鈥檚 like the fundamental question to making tutoring work. I mean, a lot of the leading tutoring researchers have said tutoring is fundamentally a question of logistics. You know, anyone, you know, so many providers have great training and many have great coaching and various things, like can you get the logistics right or can you not, know?

And so we go very, very deep with our school district partners. We have, you know, a world-class, we call them our program team, and they go in and they work because it is not like, they鈥檙e not district-wide answers to how you solve that question. You cannot go to, you know, XYZ Public Schools with 100 schools and say, OK, we鈥檙e going to do once during these blocks in this space in the school, because every school has slightly different requirements, different spaces, you know, different people. It鈥檚 all different. And so we go school by school. We help them map out a schedule. Will this instruction happen in the back of the classroom? Will it happen in a little annex room that鈥檚 right near the classroom? Does it happen in the hallway if the hallways are quiet? Often in elementary schools they are, it just sort of, again, depends very much on that school. Oh, you know, here was this space that was used for teachers to congregate, but teachers aren鈥檛 actually congregating there, or, you know, the lunchroom is actually empty from 8 to 10 AM every day, that鈥檚 unused space, like, let鈥檚 take advantage of that.

Customized Solutions for Every School

Matt Pasternack: We have worked with so many schools by this point, we just have thought through all these different permutations. And so we sit down with the schools, school by school, come up with a customized solution for each one. And, you know, the challenging part, if you鈥檙e listening to this, is that it sounds hard and expensive and, you know, and it is, I don鈥檛 want to minimize like that is, that slows down scale a little bit when you have to kind of do hard things. On the flip side, we have never come across a school that can鈥檛 do this. So sometimes you have a solution that says, oh, well, when we rewrite the language of schooling, when we completely change what schools look like and how they鈥檙e structured and what the day is, then that will unlock AI and unlock everything, and then kids can finally learn. And if you鈥檝e been around schools for a long time, I think many of us would be pessimistic that you鈥檙e going to see rapid changes in those dimensions anytime soon at scale. And so we鈥檙e really proud of the fact that we can go into any school district, any school building, and we will find a way to, you know, to build up the schedules. And the neat thing again about the application of technology, in our earlier days, we, you know, back in the kind of Google Meet and Google Sheets days, it took so much training and coaching on how to make those very lightweight technical tools work that realistically someone giving instruction could only give it to, you know, would have to give it to at least 10 or 15 or 20 kids because the investment you put in and teaching them the ropes made it, it just wasn鈥檛 worthwhile to teach a single kid.

But as we鈥檝e amped up our software and amped up this use of AI, you know, we can realize this vision. You have high school kids coming in and doing it. The school secretary, does she have a free 15 minutes or 30 minutes? That鈥檚 2 kids right there. In many schools, the principals say, hey, I don鈥檛 want to just administer this. I want to work with the you very hard, not hardest, like most difficult kindergartner, the most challenged kindergartner, the one who maybe has, you know, home circumstances that are you know, the most or, challenging, you know, who has you know, a, a learning difference, or for some reason is really struggling in the classroom. I want to take that kid under my wing as a school principal, and I you know, I want to teach that kid to read, and I want the school to see that I鈥檓 teaching that kid to read. And it goes a little bit back to, you sort of think of Steve Jobs, he鈥檇 go into Apple devices and he鈥檇 look at the wiring configurations in the background, like the things customers could never see. And he鈥檇 say, if the wires are messy in the back, then it shows that we don鈥檛 actually care about what we鈥檙e building and we鈥檙e never going to build good stuff, even though customers would never see that.

And I think it鈥檚 that attention to detail. And historically, principals maybe are really strict about kids lining up, are really strict about like certain small details that then create that school culture. Well, here鈥檚 another detail that shows, you know, my focus is on the kid, and it鈥檚 really beautiful to watch.

Diane Tavenner: Yeah, well, and thank you for that. In early elementary school, I mean, Michael and I have talked about this often, like we are hard-pressed to think of something more important than what happens in early elementary school than literally every child learns to read. And so I鈥檓 glad to hear that people are doing what鈥檚 necessary and that they understand that it鈥檚 totally possible to organize school in a way that every child will learn to read. That feels so critical. So thanks for the inspiring story of what you鈥檙e doing and the connection between the humans and the AI. Really, really fun.

Diane Tavenner: And so now, of course, Michael and I are very curious to hear what you are personally reading or listening to or watching outside of your work. We try to stay outside of our work.

We break our rule often, but if there鈥檚 something that you have to share, we鈥檇 love to hear it.

Matt Pasternack: Oh, well, no, I鈥檓 happy to. Actually, yeah, I feel like for many years between raising kids and having intense jobs, I really didn鈥檛 find much time to read other than, you know, sort of the newspaper and things like that. But I have recently joined some book clubs and gotten back into reading, which feels great, and I cherish it. I鈥檓 currently in the middle of Demon Copperhead. I don鈥檛 know if you鈥檝e read that, but the first part was so depressing. It just, you know, it鈥檚, you know, lightly based on Dickens, and it just felt like, oh my God, like just kind of going down that, going down that hole. And it was a little challenging to get through. And now things have turned around a little bit.

So I haven鈥檛 finished it. I鈥檓 excited to see what happens. But I鈥檝e been having a lot of fun with it. And then for watching, I just got to watch, I watched Alex Honnold鈥檚 ascent of the Taipei Building on Netflix.

Michael Horn: Very cool.

Matt Pasternack: Which was very fun. I watched it after it was over, so I knew at the beginning he was going to stay safe. But I got to watch it while running on a treadmill, which is a really fun experience. I鈥檓 not sure I鈥檝e ever been able to push myself that hard. It鈥檚 like, well, this guy鈥檚 doing something much harder, so I can at least run fast. So that was enjoyable.

Diane Tavenner: That鈥檚 awesome. Well, Michael and folks who鈥檝e listened for a long time will know that when I tell you we鈥檙e preparing for a trip to Morocco and southern Spain, that means that I鈥檝e got a combo fiction and nonfiction reading list that I鈥檓 working through because that is sort of how we do vacationing. And on this one, I鈥檓 digging into sort of religion, culture, history that is not very familiar to me. So it鈥檚 been a fun learning journey. I鈥檓 grateful to Gemini, who is my study buddy for this one and is actually such an incredibly useful tool for these purposes. At the moment I鈥檝e got 3 books going. So, one is called Dreams of Trespass: Tales of a Harem Girlhood by Fatima Marisi. Sorry, butchered that one.

And it鈥檚 not what you think. I鈥檓 learning a ton. It鈥檚 actually quite an interesting feminist, story,, and then 2 others. So Islam by Karen Armstrong and No God But God by Reza Asad. And those are really interesting different perspectives on the religion that I鈥檓 sort of reading side by side with each other. So super fun.

Michael Horn: Very cool. Very exciting. I always love when you share these, Diane, because as listeners also know, you鈥檝e changed my own practice around travel, uh, to start to do this habit as well, and Matt, I liked your Dickens, reference because it connects in an odd way for the one that I鈥檓 going to do, which is Diane knows I don鈥檛 read a ton of fiction, but I actually finished a fiction book here, In the Shadow of the Greenbrier by Emily Matchar. I鈥檓 probably also bungling her last name, but I picked it up, honestly, I was at synagogue. I saw it in the temple library, and Greenbrier was a place that we used to vacation with my grandparents and my cousins a few times growing up over Christmas. And so I was sort of curious, and it鈥檚 like this intergenerational Jewish sort of mystery story, if you will, trying to piece together different puzzle pieces. And the Greenbrier is sort of the central part of it.

But the only reason I say that, Matt, is I remember one of the years that we vacationed at the Greenbrier, they had like the foremost Charles Dickens expert or something like that in residence, and he gave lectures, which we as 12 or 13-year-olds dutifully attended and I think did not cut up too much during, which, which was impressive for us. But, so I, I feel like I鈥檓 connecting on a bunch of these things at the moment, but, uh, it was a fun read, and Diane, one of the reasons I don鈥檛 read a lot of fiction, I forgot, is because when I read it, I don鈥檛 put it down and I become a bit of a zombie around the house for a couple days.

Diane Tavenner: So yeah, there is such a thing as binge reading, just like binge watching, right?

Michael Horn: Exactly. And nonfiction, while I love it and I read it a lot, turns out it doesn鈥檛 do that for me, whereas fiction I鈥檓 a lost cause around the house.

Matt Pasternack: The secret is just simultaneously reading it on the Kindle and Audible, and then you can, you know, you can volunteer, hey, I鈥檒l go do the dishes, you just put in

Michael Horn: Exactly, plug it in and power through.

Matt Pasternack: You keep going, right, you know, right where you were.

Michael Horn: So good, good tip, good power tip. All right, huge thank you, Matt, for joining us. And again, to all of the listeners, who keep coming in with all sorts of feedback, both positive notes, questions, we like those and then some of the hate mail too, we love it all because we learn tons, and just, keep it coming off this conversation. We look forward to more, and we look forward to seeing you next time on Class Disrupted.

This episode is sponsored by LearnerStudio.

Applications are expected to open in 2027 for the Khan TED Institute, a joint project of Khan Academy, TED 鈥� the purveyors of the popular TED Talks 鈥� and the Educational Testing Service.

鈥淚 think there’s always been, frankly, some need for a program like this,鈥� said Khan Academy founder Sal Khan. Many people, he said, can鈥檛 afford a college degree or can鈥檛 take the time out of their work lives to attend four years of classes. 鈥淚t could be that they have pursued a degree, but it’s not giving the signal that would give them the opportunities that they would want.鈥�

Another founder, Amit Sevak, who leads ETS, acknowledged that they are still working out many of the details, but that the new institution could someday enroll 鈥渢ens of thousands鈥� of students, rivaling flagship state universities. Sevak said he鈥檚 鈥�100%鈥� anticipating that its instructors will be humans, most likely a large network of adjuncts.

鈥淲e still believe in the value of a human teacher,鈥� he said. 鈥淲e think that there’s so much socialization and collaboration that takes place [in the classroom]. There’s also the classic need for classroom management and some pedagogical oversight over the assessments.鈥�

Here are five things you need to know about the new enterprise:

1. It鈥檒l offer a bachelor’s degree in applied AI in various fields such as business, marketing, human resources, healthcare and more.听

The college will offer a full undergraduate bachelor’s degree organized around three pillars: core academic knowledge 鈥� math, statistics, economics, computer science, science, history and writing 鈥� applied AI skills and 鈥渄urable鈥� human skills such as communication, leadership, collaboration, peer tutoring and public speaking. 

Early employer partners include Microsoft, Google and , an AI app development site.

2. It鈥檚 expected to be competency-based, cost less than $10,000 and take as little as half the time of a traditional bachelor鈥檚 degree.

The college鈥檚 founding partners say its total cost will likely be under $10,000, a fraction of the of a four-year degree.

Rather than requiring four years of seat time, Sevak said, the institute is built around a competency-based model, offering students the opportunity to advance when they demonstrate mastery. That means students could potentially complete the degree in two to three years, he said, depending on how quickly they demonstrate required competencies.

That opens it up to many different kinds of students, he said, including motivated high schoolers who want to earn undergraduate credits quickly before graduation, working adults seeking advancement in their jobs and students already enrolled in traditional colleges who want to stack an AI credential on top of their existing undergraduate credits.

Khan said the new college 鈥渋s something I鈥檝e thought about doing in some way, shape or form, for many years, and the changes within the job market, because of AI, only accelerated that.鈥�

He said the idea came out of conversations with TED chairman about a year and a half ago. 鈥淲e started saying, 鈥業t feels like there’s something powerful between Khan Academy and TED. We’re both learning organizations. Khan Academy is known for academic learning from K-through-14. TED is known as [embodying] lifelong learning. And it’s about human connection. And it feels like we both have fairly unique brands in the not-for-profit space and the education space.鈥欌��

Khan later spoke at an ETS trustees dinner and got to know Sevak.

鈥淭hey’ve been looking at the same things,鈥� he said, 鈥渁nd they’ve also come up with a framework on durable skills and thinking about ways to assess them. And we realized, 鈥楲ook, the world needs this. And if the three of us come together, this will be very credible and hopefully has a high chance of helping a lot of people.鈥欌��

3. It鈥檚 an 鈥淎I-first鈥� institution, weaving artificial intelligence into how courses are designed, taught and assessed.

Sivak said courses will be shaped by AI and teaching will be supported by AI agents, software systems that can tutor students, answer questions and provide feedback. And students will be prepared for work in 鈥淎I-native鈥� environments.

Instruction will likely be 100% online at the college鈥檚 launch, with an emphasis on asynchronous coursework to accommodate students in different time zones and life circumstances. Over time, Sevak said, they鈥檒l likely explore a hybrid format.

4. Khan Academy will provide the college鈥檚 learning platform and pedagogical infrastructure, despite its founder鈥檚 tempered enthusiasm about AI and learning.

TED, the conference organization best known for its short, , will incorporate its content into the curriculum, giving students access to live talks, Q&A sessions and community-based learning with TED speakers.

And ETS, the testing and measurement organization that produces the GRE and TOEFL tests, will contribute its assessment expertise, said Sevak.

Khan Academy, the popular free tutoring website, which has about and operates its own , will offer its technology to deliver the college鈥檚 coursework, organizers said. Khan, who founded it in 2008, will hold the title of 鈥淭ED Vision Steward鈥� in the new partnership.

The announcement comes just a few days after Khan told Chalkbeat that the learning revolution he predicted in 2023, upon Khanmigo鈥檚 release, .

In September 2022, Khan and Kristen DiCerbo, the organization鈥檚 chief learning officer, were among the first people outside of Open AI to get access to GPT-4, the large language model that at the time powered ChatGPT. Their experiments gave rise to a revolution in Khan鈥檚 thinking: In 2023, he delivered a TED Talk in which he predicted 鈥渢he biggest positive transformation that education has ever seen,鈥� saying we鈥檇 soon be able to give 鈥渆very student on the planet an artificially intelligent but amazing personal tutor.鈥�

In 2024, Khan鈥檚 book, , bore the subtitle 鈥淗ow AI Will Revolutionize Education.鈥� 

But more than three years after Khanmigo鈥檚 launch, Khan admitted, 鈥淔or a lot of students, it was a non-event. They just didn鈥檛 use it much.鈥�

A few students, he said, have used the AI chatbot readily, while others haven鈥檛. AI tutoring, he concluded, doesn鈥檛 necessarily motivate students to learn or fill in knowledge gaps they need to learn more. He鈥檚 still optimistic about AI in education, but also sees its limits. 鈥滻 just view it as part of the solution,鈥� he said. 鈥淚 don鈥檛 view it as the end-all and be-all.鈥�

On Monday, Khan told 社区黑料 that AI is 鈥渏ust going to be part of our arsenal to help make more engaging tools. Maybe we鈥檒l be able to give more rich assessment practice. Instead of having multiple-choice questions, you can start to have 鈥榚xplain your thinking鈥� [questions]. So it starts to open up the aperture.鈥�

5. It鈥檚 very much a work in progress.

Speaking four days before the launch, Sevak admitted that nearly everything about the venture 鈥渋s still evolving,鈥� and that the team is 鈥渨orkshopping the pedagogical design鈥� of the new college.

Sevak said the institute is in talks with regional and national organizations that can offer 鈥渢he highest form of accreditation,” a step that would set it apart from a growing number of online certificates, micro-credentials and boot camps. 

鈥淲e’re really in the early days, and it’s just going to take some time for us to adapt,鈥� he said. 

The college鈥檚 curriculum isn鈥檛 yet finalized and applications are 12 to 18 months away. Likewise, the specific structure of its hybrid and asynchronous models, its faculty roster and the full range of majors are all still in development.

鈥淥ur intention is, over time, to have a whole range of specializations,鈥� said Sevak. But the program鈥檚 core is designed to prepare students 鈥渢o be really AI-centric鈥� for a new reality. 鈥淲e’re seeing [AI] as ripping through the economy,鈥� creating a lot of uncertainty for young people. 

More to the point, said Khan, 鈥淲ork is changing very fast. AI is changing everything.鈥�

Compared to a , they鈥檙e less excited and hopeful about the change it could bring and more angry at its existence, citing concerns about AI鈥檚 impact on their cognitive abilities and professional opportunities.

Respondents said they used AI at nearly the same rate they did before 鈥� they reported only a slight increase in daily and weekly exposure 鈥� but when asked how it makes them feel, the answers revealed growing misgivings. 

Thirty-one percent said it made them angry, up 9 percentage points from 2025. And just 22% said it made them feel excited, down 14 percentage points from last year. Only 18% of respondents said it made them feel hopeful, marking a nine-point drop. Forty-two percent said it made them feel anxious, roughly the same as last year. 

Zach Hrynowski, senior education researcher at Gallup, said the switch was swift. 

鈥淥ne of my working theories is that (it鈥檚) the high schoolers, who are in their senior year, or especially those college students, who are maybe thinking, 鈥楢I is taking my job. I just went to college for four years: I spent all this money and now it’s turning my industry upside down,鈥� he said. 

Only 46% of respondents believed AI would help them learn faster, down from 53% the prior year, Gallup found. Fifty-six percent of respondents said it would help them to expedite their work compared to 66% last year. 

Hrynowski notes, too, that users’ unease wasn鈥檛 entirely tied to the amount of time they spend engaging with AI. 

鈥淵ear over year, among that super user group, they’re much less excited, they are much less hopeful 鈥� and they are more angry,鈥� he said. 鈥淪o this is not a case of some people who are adopting it and loving it and some people who are just avoiding it and feel negatively about it.鈥�

Nearly half of respondents said the risk of the technology outweighs the benefits in the workforce. Just 37% believed it would help them find accurate information, down from 43% the prior year and only 31% believed it would help them come up with new ideas compared to 42% in 2025. 

The survey also notes some disparities by age and race. For example, older Gen Zers are more likely than younger ones to voice concerns about AI鈥檚 impact on learning in general. 

Asked how likely is it that AI designed to mainly complete tasks faster will make learning more difficult in the future, 74% of K-12 respondents said it was 鈥渧ery likely鈥� or 鈥渟omewhat likely鈥� compared to 83% of Gen Z adults who said the same. Men and Black respondents were also less concerned about learning impact than their peers overall.

Results are based on a survey of 1,572 people spread throughout every state and Washington, D.C., conducted between Feb. 24 and March 4, 2026. It was commissioned by the Walton Family Foundation and , Global Silicon Valley. Together, Walton Family Foundation and Gallup are conducting ongoing research into Gen Z’s attitudes toward AI.

Hrynowski believes there might be a link between recent revelations about the harmful nature of social media and AI-related distrust: Many of the respondents came of age, he notes, just as former surgeon general Vivek H. Murthy called for a about its use. 

shapes the user experience in social media. Just last month, a California jury found social media company Meta 鈥� owner of Facebook, Instagram, WhatsApp, Messenger and Threads 鈥� and YouTube injured a young woman鈥檚 mental health by design in that could encourage untold others. 

This was the second of two critical decisions: Just a day earlier, a New Mexico jury found Meta 鈥� and hid what it knew about child sexual exploitation on its platforms.

鈥�I’ve always been very impressed from the start of this work with Gen Z that across the board, not just with AI, they are keenly aware of the risks of technology, whether it’s social media, whether it’s AI or screen time,鈥� Hrynowski said. 

They are not the only generation to harbor these worries. A growing number of parents of K-12 students are pushing back on their screen time, not just , but  

Despite respondents鈥� skepticism about AI, they鈥檙e also readily aware that the technology won鈥檛 be walked back: 52% acknowledge that they will need to know how to use AI if they go to college or take classes after high school, while 48% think they will need to know how to use AI in the workplace.

An earlier Gallup study, released just last week, shows 42% of bachelor’s degree students have reconsidered their major because of AI.

Gen Z, in its reluctant acceptance of the technology, wants help in how to navigate it, both in an academic setting and in the workplace. Schools are stepping up, the survey revealed: The share of K-12 students who say their school has AI rules moved from 51% in 2025 to 74% this year.听

Disclosure: Walton Family Foundation provides financial support to 社区黑料.

In the latest episode exploring new school models powered by artificial intelligence, Summit Public Schools鈥� Cady Ching and Dan Effland join Michael Horn and Diane Tavenner to discuss Summit鈥檚 transformation into an AI-native school model. The conversation examines how clarity around school outcomes and model design enables the effective integration of new technology, followed by insights into the evolution of Summit鈥檚 expeditions. Ching and Effland emphasize the importance of a holistic, purposeful education, as well as the need for a robust technology infrastructure to scale innovation.

Listen to the episode below. A full transcript follows.

Cady Ching: I think what has been really helpful for me is to list the ways that a model is not. It’s not a curriculum, it’s not an LMS, it’s not a schedule by itself, it’s not a set of beliefs or a graduate profile by itself. Those are parts of a model, but a lot of the building that we’re seeing right now is focused on building for parts versus building for an actual whole model. And so the AI-native model is how all of those model elements are working together. And it is not going to be replacing a school model. It’s going to expose whether or not you actually have a model. And I think AI is forcing a lot of school systems right now to get really honest, because if you don’t know what students are supposed to be learning and you’re not sure how they’re showing that or what adults are responsible for, AI just layers on complexity and, quite honestly, chaos. But if you do have the level of clarity of what Dan is speaking about, AI is actually making systems work a lot better, or it can make systems work a lot better.

I think the jury is out on the tools that we need and how we can create the tools that we need. But AI really isn’t replacing, it’s revealing whether or not your school model actually exists.

Diane Tavenner: Hey, Michael.

Michael Horn: Hey, Diane, it is good to see you with some excitement for today’s episode.

Diane Tavenner: Yeah, we have a real treat today. We’ve got two of my favorite educators in the world joining us for what I’m sure is going to be just a really interesting conversation.

Michael Horn: Well, and for years, as obviously I’ve learned about Summit from you, direct from you, and yet it’s been nearly 3 years, I think, since you passed the baton, if math is still a thing. And I know from afar that the team continues to be among the most innovative schools in the country and so I know that they continue to think about reinvention, and frankly, you know, what does Summit need to look like? How can it get even better? All these questions for its learners. And so I’m incredibly excited to dig in and learn about what they’re calling Summit 3.0 on today’s show. I will say it’s also interesting to have this conversation because we’re sort of in our model geek out, if you will, at the moment, right? While we’re having this conversation, we’ve had the founders of Alpha School, Flourish on, both of which are designed as AI-native models. And for those who listened to those episodes we sort of created a little bit of a side-by-side, if you will, where we said, hey, Summit is here as this baseline for a pre-AI model trying to do personalization or optimization of each kid’s learning. And we explored what can you do in an AI-native world? How can you design differently? But today what’s exciting, I think, is we’re going to get to dig into what does it look like for an existing model with that orientation to become, quote unquote, AI-native.

And as you know, transformation and how organizations reinvent themselves, that’s something I get really passionate about and excited. So I cannot wait to learn from the real-life example in progress.

Diane Tavenner: Well, we’ve got the two perfect people for that conversation, Michael. And so let me introduce you to Cady Ching, who is the CEO of Summit Public Schools, where she was an extraordinary teacher and school and network leader for a decade before taking on that role. So she brings this full spectrum of experience to this next phase. And Dan Effland, who is the senior director of innovation at Summit, where he was also an extraordinary teacher and school leader before taking on this new role of leading for the second time in the history of Summit, the reinvention of the model. And so welcome, Dan and Cady. We’re so happy that you’re here with us and excited to talk to you about the work you’re doing.

Cady Ching: Thank you. Thank you so much. I’m excited too. It’s coming at this moment for Dan and I where we’ve been trying on a lot of language about where we’ve been, where we are today, and where we’re going. So selfishly, this is a milestone for us.

Michael Horn: Well, and I get to feel like I’m jumping in on a team huddle of y’all. Yeah, this will, this will, this will be fun.

Cady Ching: Welcome, Michael.

Michael Horn: Thank you.

What Is a School? 

Diane Tavenner: Dan and Cady, a few weeks ago we got together and you walked me through the thinking and planning you’re doing. And honestly, I was captivated, you know, because I got stuck on it and I wanted to dissect every word. By this simplest definition of school, it’s honestly the simplest definition I’ve ever read of a school. And I wanted to start there today because I really think we always have talked about getting to the simplicity on the other side of complexity. And I think you’ve done it with this definition, and I think it’s going to be really powerful in this next chapter. And so maybe, Dan, kick us off. And if you will share that definition and a little bit about how it came to you or how you all came to it in your process and what you think it unlocks.

Dan Effland: Yeah, happy to. And thanks for having me here. I’m so excited to talk to you all. Yeah, so, I mean, we’ve been working on this for years, right? What is simplicity on the other side of complexity? And I think as we’ve been digging into what does redesigning look like, it became really clear that you have to get down to some foundational elements to avoid designing within conventions and not even really realizing you’re doing it. And so the way we’re thinking about schools is simply, it’s a group of young people. It’s a set of outcomes or competencies. And then it’s a set of resources that help you support young people to achieve those outcomes or competencies. That’s it.

Kids, outcomes, resources. And stripping all the way back to that has allowed us then to engage with our community, because all this work is like with students, caregivers, and educators, and go like, OK, what do we really want? What do schools really need to be? With full freedom, we call them dreaming sessions, where we can really engage off the simplest foundational elements and not get hooked by any of the conventions that have existed, you know, for decades or longer than that in a lot of cases.

Summit 2.0: Evolution and Vision

Michael Horn: It’s really cool because you’ve sort of, like you said, you sort of have a conversation around what those end posts, and we can sort of figure out what’s inside the box to get there apart from what’s always been there. But before we go to that sort of Summit 3.0 vision and where you’re thinking currently is, because I’m imagining you’re going to have lots of trade-offs and changes as you go through the design process, but I think it would be helpful to do a quick turn on Summit 2.0. Both to ground, frankly, our audience, but also to set up a question of how things are changing and where and so forth so that we can understand that. And so I’d love, and maybe Cady, you dive in on this first, how would you describe the Summit 2.0 model, which was not only in your schools, but schools across the country? It’s one of the reasons I think it can be called a model,  it’s scaled beyond Summit itself, right? And as you think about that, the new model, what is it in the Summit 2.0 that you’d say, we really want to hold on to this? Or where are the things that you’re saying, hey, actually, that’s something we can leave behind or start to question whether we want to change that?

Cady Ching: Yeah, thanks for asking this question. I think it’s so important. The reason why I keep smiling when you all say Summit 2.0 and 3.0 is because Dan and I actually got into it a couple weeks ago about if we wanted to use that language or not. And my issue with it was I think it’s really, it serves a purpose because like to Diane’s point, it is simplicity at the other end of complexity. And there is a danger in the simplification of the 2.0 and 3.0 because at Summit, we really think about innovation in two ways. One just being innovation through refinement, which is the day-to-day tightening of the model elements that we’re building on for these larger moments of innovation, which we call innovation for redesign. And so those are sort of the sector-shifting, big model, what we call Big M changes. But I’m going to use Summit 2.0 and 3.0 language today in shorthand.

Michael Horn: Thanks for doing it for the listeners.

Cady Ching: Yeah, and so Summit 2.0 really speaks to our personalization era at Summit, where we showed personalization doesn’t need to be a luxury. And we did that by designing cohesive student and teacher experience., and it included model elements like mentoring and skills assessment and differentiation using real-time data, which we enabled through tech. And the tech that we co-built was called the Summit Learning Platform. For me, what I think was most remarkable about what we proved in Summit 2.0 is what you mentioned. It was scalable, and it did scale, and schools were able to implement and sustain the Summit model on public dollars. Which was remarkable. And so we reached 100,000 students, 6,000 educators, and 400 schools across 40 states.

And we did it with district, charter, private, rural, suburban, and urban. It was completely shifting the field. And then we normalized mastery-based learning, personalized playlists and skills and habits in a way that now is the foundation and the baseline in so many places that we’re now talking about building these AI-native models on top of. And so to the second part of your question, which I’ll kick off and then, Dan, I’m going to pass it to you to add on, we think about model elements and processes that we want to carry forward into Summit 3.0. In the process side, which is where I thrive, we were successful because we were leading from this intersection of the learning science, community engagement, and technology, and we centered teachers and students at every part of the design.. And we’ve used those same design principles to continuously improve our model since Summit 2.0. For me, I feel like we’re 4 years into Summit 3.0, and we’ve already gotten some really exciting data back about situating us as leaders in the field again around what we’ve built on top of the personalization.

In last year, this is our most recent data, we saw that our Summit alumni have some of the highest post-graduation incomes and lowest debt loads, as compared to other top-performing charters. And this is the type of longitudinal outcome evidence we’ve been really longing for. And when you think back about how Dan just defined the system, what that data does for us is it grounds us in that we do have a really strong set of outcomes and competencies that are timeless. Our young people are now achieving them, and we’re letting go of the old technology to create space for AI-reimagined infrastructure that’s going to help us to better allocate resources. And we think our biggest resource levers are people, technology, and time. So that’s really how we’re thinking about Summit 2.0 setting us up for Summit 3.0.

Michael Horn: Dan, did you want to jump in there and add some?

Dan Effland: Yeah, yeah, I think I’ll just like, you know, I think, you know, Cady and I were both teachers in Summit 2.0. We were both school leaders in this, and so we have a lot of really direct connection to it. And the thing that really makes me think about it is like, you know, the learning platform is no longer in existence, but the elements of the model really deeply took root. Mentoring, mastery, what we called habits of success, I think we’re calling durable skills in our world now. Like, I’m fine with it, whatever we want to call it. It’s become ubiquitous. And I think it really helps. I mean, I think it really gives us a sense of a strong foundation of like, we’ve done this before, we’ve built a model that’s scaled and really stuck.

And it doesn’t matter if the technology, you know, is stuck or not, because that technology is not the model. The tech model is these elements of how you support kids to master these outcomes with whatever available resources you have are. And so, yeah, I think there’s a point of pride when we think about, you know, what we’re begrudgingly calling Summit 2.0. And then I think there’s a sense of the strength of the foundation to then build what’s coming next.

Personalization & Durable Skills

Michael Horn: It’s interesting. And we’ll come back to the technology, I know, and we want to circle back to that. But hearing Cady, you described the model, used a few words that I think are really important for people to hear. One of them was cohesive, because I think a lot of the tech efforts right now around personalization in so much of the country are the opposite of cohesive. And that’s why we’re seeing a blowback sometimes against technology, because it’s sort of all over the place and hundreds of things going on at once for a young person with tons of distractions. And you talked about it being grounded in the learning sciences and personalization as a, as a means, not the ends, right? And, and then you have these longitudinal outcomes. And I’m just calling them out because I think people often lose sight of, this is the bedrock, right, of how we build from, and then go from there. And the other piece, and Dan, you just referenced this, the field is now calling it durable skills.

I still prefer habits of success. Let me just be on record on that one. But one of the things you all really did well around Summit 2.0 was have incredible clarity on the mission, what success looks like, such that you could measure in the way you just said, Cady. And I didn’t know those stats. I mean, it’s fascinating., and then you had these commencement-level outcomes, right? You were super clear on what does it look like from a, you know, for a Summit graduate as they go out in the wild. And it seems in some ways those commencement-level outcomes have been precursors to the movement across states that we’ve seen in the Portraits of a Graduate. And I do think that there’s some key differences. I’ll hold my editorial back on what those are more because I want your take on that.

Like, what, if anything, are the differences and, and between those commencement-level outcomes that you all have defined, the portraits of a graduate that we see states doing, and more broadly, like, what’s the importance of being super clear on what those outcomes are and, and how you’d know, on the other side, if you could speak to that. And I don’t know, I’ll make it a grab bag of which one of you wants to jump in on that.

Dan Effland: Dan, take it away. Awesome. Yeah, I mean, so our vision has been the same for 23 years. It’s preparing young people for a fulfilled life, really all people. We think of our staff as part of that too. And fulfilled life is in some ways, again, simple. It is purposeful work, financial independence, strong community, strong relationships, and health. And so that’s given us a holistic picture, a holistic point B that we’re always going for.

You know, I don’t, I don’t know how I compare it to Portrait of a Graduate or Portrait of a Learner. What I know is it gives us a lot of clarity in that you can’t design a coherent model without clarity of where you’re headed. And that it’s also really important that that clarity is holistic and is not simply a set of academic outcomes. It is much broader than that. And that gives us a huge advantage in this work right now because we’re not spending a lot of time. We certainly talk to our community and affirm, you know, on a regular basis, is this still what people want? Is this still what our communities are after? And it is. And so we can move right to like, okay, how do we get there?

Cady Ching: The thing that I would add on top of that is, I loved, Michael, what you called out around the language of a model. I think that at the operator level, and when I’m talking to, to other school leaders, this word is used in a lot of different ways. And I think what has been really helpful for me is to list the ways that a model is not. It’s not a curriculum. It’s not an LMS. It’s not a schedule by itself. It’s not a set of beliefs or a graduate profile by itself. Those are parts of a model.

But a lot of the building that we’re seeing right now is focused on building for parts versus building for an actual whole model. And so the AI-native model is how all of those model elements are working together, and it is not going to be replacing a school model, it’s going to expose whether or not you actually have a model. And it’s, I think AI is forcing a lot of school systems right now to get really honest, because if you don’t know what students are supposed to be learning, and you’re not sure how they’re showing that, or what adults are responsible for, AI just layers on complexity and quite honestly, chaos. But if you do have the level of clarity of what Dan is speaking about, AI is actually making systems work a lot better, or it can make systems work a lot better. I think the jury is out on the tools that we need and how we can create the tools that we need, um, but AI really isn’t replacing, it’s revealing whether or not your school model actually exists.

Diane Tavenner: I鈥檇 love it if we go back to your simple definition, Dan, that we started with, when we sat down. You use the word package of outcomes, and I was obsessed with that word package for this reason, because you know, maybe I will jump in here a little bit on the portrait of a graduate. 

Michael Horn: The table’s been set for you, Diane. 

Diane Tavenner: Yeah. And one of our, you know, Summit’s longtime beloved board chair, board member, who honestly is one of the most forward-thinking, I think, philanthropists who launched a scholarship for Summit graduates going into Pathways years ago, like ahead of the curve, you know, sent us a note the other day with a real critique of portraits of a graduate. He was sort of reading about them and was just very, you know, like, what are these people thinking? And I think what he was responding to was a lot of the portraits of the graduate, like, feel very checkboxy and compliance-oriented. Versus this sort of holistic. And I know that’s not the way they were intended.

AI Evolution in Education Models

Diane Tavenner: They all have good intentions behind them, but the way they have been sort of brought to life and then communicated and then implemented are what Cady, I think, is speaking to, not as a model, but as like these individual components that don’t have a coherence about how they’re actually organized an organized set of resources to achieve those package of outcomes, if you will. And so I think that what you all just described is at the core of your success going forward and what an advantage you have. And it really speaks honestly to the durability that you’re carrying all of that forward in this next phase, that being, living a life of wellbeing it actually hasn’t changed, right? The elements of that haven’t changed, and that’s what you’re equipping young people for. So, you know, in a recent episode, Michael and I had a conversation, just the two of us, which was super fun, and we were dissecting a way of thinking about school models in three buckets. And I know you are both familiar with this framework, which is essentially that, you know, Model 1 will use AI to make sort of the existing industrial model school more efficient and better. Model 2 will stretch the bounds of that industrial model school with integrated AI. And Model 3 will be AI native, you know, essentially built from the ground up with AI capabilities that are assumed to be at the core. And, you know, as you think about where you’re now going with Summit 3.0, how do you view it in the context of this framework? And, you know, what does AI make possible that wasn’t possible in 2.0 because it was designed pre-AI?

Dan Effland: Love this question. And I did listen to that episode. So I’ll start with the model part, and then I really want to get into what AI makes possible and kind of what it pushes us to do. So I love reading like Learner Studios’ 3 Horizons model. I love Bob Hughes’ paper on the 3 models. I find that stuff really, really important for evaluating what exists and really valuable for visioning and for getting into this place of what really is possible. And I think, and that’s really useful. I will say, when we start designing and working with our young people and working with our caregivers and our educators, I actually find it useful to kind of set those categories aside and to ask the more foundational questions around, like, we know where we want to go, we have this clear vision, we have this really simple, you know, conception of what a school is with kids’ outcomes and resources.

And now let’s go from here. And when you get into, like, as we’ve talked about, we have a lot of clarity about our outcomes already. We really believe deeply that this holistic model of a healthy, thriving, you know, young person, young adult, adult is going to be durable regardless of the transitions that are happening in our society. But when it comes to the resources part, now we have this whole huge different potential, one, AI being a resource, but also a way that I think we’re most really interested when it comes to AI is how we can use it if we integrate it into our tech stack. Really how, like, with a really robust knowledge graph and really strong data layer, you could be dynamically reallocating resources in a way that just would be impossible for people. You know, like when I used to build an annual schedule, like the primary schedule with our Dean of Operations, she and I would sit in an office for a week with a spreadsheet to make a schedule for the year that never changed, right? Like, it’s just so labor-intensive. But now I think when we think about AI as part of our infrastructure, and it’s kind of a layer in our tech stack interacting with a really robust knowledge graph and data layer, we can start to ask ourselves, like, how do we get the right resources to the right kids at the right time for the right outcome? And really get very, very precise, and also do that dynamically. And I think that then allows us to think about personalization, just-in-time instruction, integrating real-world experiences, ensuring that personalized learning still happens in community and there’s deep human connection that is part of personalized learning journey in a way that was, was not possible when, you know, 12 years ago when we were thinking about Summit 2.0, the technology just didn’t exist.

And so, I mean, it’s exciting. I mean, I really think there’s incredible possibility there. And while there’s definitely lots of really cool tools being built, we’re much more focused on the, like, where does this fit as part of our technology infrastructure or our tech stack, because we think that’s, like, potentially a huge lever for transforming learning for young people.

Current Applications of AI in Schools

Michael Horn: It’s fascinating to me, ’cause you just named a number of things that AI could do that I had never thought about in terms of, like, dynamically changing the schedule for, you know, the school and students and, like, there’s some pretty cool things you can start to imagine that ripple out of that. One of the things in that conversation that Diane referenced that she and I agreed to hold ourselves accountable for was to get really specific when we talk to school leaders about, so what’s happening today in your schools that’s actually leveraging AI or is quote, unquote AI native, if you will? And so you all are obviously still in the design phase for 3.0. I use that with trepidation now, but put that aside for a second. Like, today, if I were to, you know, get to be in California again and I was hanging out in your schools, what would I see that’s powered today by something that’s AI native? What is it? What are the tools? What does it look like? What does it do? What are you building versus partnering with? Give, give us a sense of some concrete applications. Anywhere in the tech stack or during the day, that is AI-powered?

Cady Ching: I think this would be a good opportunity to talk about a specific tool that we’re using, which maybe not ironically is Futre as one model example of what it can look like. And Dan can speak to specifically what it’s looking like in the student and teacher experience. But one of the reasons why I start with speaking about a specific tool is because I think that largely edtech has not鈥� has been really unsuccessful in solving for what we need to operationalize innovative school models. And Futre has been a nice shift of pace for us because it is truly a tool that is building for the child versus fitting a child into a tool or larger system. And I think that the way in which we’re using it with our young people can work in many H2 and H3 model contexts because it’s able to give us real-time data about our young people and then allowing us to build their student experience based on the data that we have about them. Dan, can you introduce, Michael a little bit more to Futre and how we’re using it at Summit?

Dan Effland: Yeah, absolutely. So Futre right now we’re using with our juniors and seniors, although we anticipate starting younger, in the coming year. And right now, our juniors are really using it to do a lot of career exploration, which the tool excels at, and really like exploring very deeply different possibilities. And then what those possibilities mean as far as what they need to be working on now or experiences they have between kind of their current point A and their future point B. And then our seniors are using it to get more concrete about what really, what is my next step? What does that mean? What is the thing I’m doing immediately after high school?  鈥� I think we deeply believe this and will proudly say it is best-in-class career-connected learning. It is. Absolutely. It is the thing when we do 鈥� when I do focus groups, when we do alumni data, kind of research, it just comes up over and over again because our young people actually get out in the community or within the school building and really doing what we now are calling real-world experiences. We’ve called them lots of different things over the decades, but we are 鈥� one of the things about that though is that kind of like we were talking about, how do we really curate the journey with this resource allocation stuff? Just tracking all of those different experiences, often there’s 50 or 60 choices for students at one school when we had those expedition cycles. We’re now pulling those experiences onto the Futre platform so we can really start to map what students have been doing, what they haven’t been doing, maybe what they should be doing. And then their mentor can take an even more engaged kind of role in coaching them through that pathway. We’re really excited about that.

We’re kind of just starting, you know, to pull those on. But I think in the future it’s one of the things that we see that the Futre tool will be really, really helpful with because, you know, young people need coaching as they’re figuring out that concrete next step.

Michael Horn: So super interesting. I actually have two questions, but let me go to you, Dan and Cady, first. And then I have a question for you, Diane. I’m going to put you on the hot seat. But I think we’re allowed to do that. But it’s interesting. You just said something there in your answer, Dan, which was then the mentor or coaching.

And so just like to put a fine point on it, The, like, this works really well because you have a model where there is that function that is meeting on a regular weekly basis, right? And like, so therefore that touchpoint, like it’s coherent again to use that word, but I, I would love a quick update on how Expeditions has evolved because when I think when Diane was exiting Summit, like, y’all were in the middle of redesigning it and I’ll be super honest, like even though she and I talk basically weekly, I don’t actually know the new version of Expeditions. And so, I still have a slide in my talk about Summit that says, you know, like every 8 weeks or whatever, you go off for 2 weeks. And y’all should update us on what’s the current state of Expeditions at Summit.

Cady Ching: Yeah, I’ll respond to 2 pieces. One, with the mentoring piece, that model element does exist. One of the reasons why I personally love Futre is because it takes some of the lift of mentors needing to be the vessel of all career pathways off the human. So when we think about that resource allocation of, you know, people, talent, it’s creating a better, more coherent system for the adult as well, which has been so important because we love to center our teachers as well in the design. And then the Expeditions redesign, it’s been really cool. We’ve been, you know, continuously shifting that program based on what our alumni are sharing back with us, based on how the world is shifting. And of course, AI, as so much a part of our students’ experience today and in the future, has shifted it again. It is non-graded鈥� so this is actually surprisingly one of the most controversial things when we rolled it out to parents鈥� they are not receiving grades on the different career exposure pieces that they try out as they’re with us at either the high school levels or as early as 6th grade in Seattle.

And it’s really about ensuring our students get about 9 career exposures between the time they start with us to the moment they leave, because we know it’s really important for them as they develop their identity to see themselves in different career pathways that are all mapping towards high opportunity where they can build their generational wealth for their family. So it’s probably pretty similar in terms of the time allocation. They’re in sort of what we call their core classes for 6 weeks, and then they’re pausing for 2 weeks to go out, usually in the upper grades, off campus. You don’t see 鈥� when people come to observe this on our site, they’re not actually a lot of kids in the building because learning happens without walls. Dan, what else would you add as you’re going? Dan is quite literally on an expedition tour currently. He’s at one of our school sites right now, and right after this recording, he is going to go in and speak to our teachers. So what else would you add?

Dan Effland: Yeah, I mean, I think that’s an important side of it is so that, I mean, one, it’s just, I was still in a school leadership position when we transitioned to this kind of redesigned Expeditions, and I just can’t tell you how powerful the experiences are. I can think of so many stories, so many young people, but like one in particular that a young, he’s 鈥� well, he’s probably not even that young now, but he’s 25, but he was a young, young man at the time who was really, really struggling. And this kid was having discipline issues, attendance issues, struggling, like, not necessarily living at home on a regular basis. And we really, we thought we were gonna really lose this kid. And he started doing an expedition experience related to culinary arts. After he did that first one, he did a second one, and then there was kind of a sequence of them where he had, you know, like the first one was kind of like a survey course. It was the community college. It was about 25 kids.

Finding Passion and Purpose

Dan Effland: Then he was able to do one where he was actually kind of shadowing one of the actual culinary arts program college students and learning in a second wave. So I’m having a hard time not using his name, but I’m going to keep it out. But I just loved this kid. And he found his pathway. And not only did he find his pathway and ended up going to a culinary arts program and graduating and now works, you know, like in the culinary arts, you know, scene in Seattle, his attendance improved, his grades went up, his connections with his mentor, with his teachers, with his peers, which were, you know, fraught, got better and better. And he became a healthier human because purpose and passion and having a pathway is essential for all of us. And we’re at a time when, you know, you can read about this everywhere, there’s studies, our young people are really searching for that clarity about purpose and pathway. And when you see it, I mean, it’s just like Cady said, it’s kind of hard, like it’s not a good thing to tour because the kids are mostly out in the community.

Dan Effland: But when you have the privilege of being a school leader and you see these kids over the years and they do their cycles, you just, the impact is unbelievable. So yeah, I just wanted to, yeah 鈥�

Designing Education for the Child

Michael Horn: No, the anecdotes make these things always so much more powerful. And I mean, you can, through your story, hear him building a positive identity of himself, right? And that’s incredible. Diane, something Cady said made me think of it, which is obviously, you know, folks who listen to us know that you’re the entrepreneur behind Futre. I now understand why it was originally called Point B based on Dan’s language and I guess, but she said something interesting, which was like a lot of edtech has not helped the launch of new model design, right? Because it’s been, and that, that’s sort of been obvious to me for why, right? Because the market is schools as they are, and venture capital wants big markets, and right, like, it’s 鈥� so it’s, it’s this sort of reductivist thing that happens. But she said you’ve been designing for the child, and so you’ve been able to escape that and I wondered if you just might want to reflect on that, because I imagine it is still hard though, um, because you’re still like 鈥� schools are the conduit to the kids. So just sort of like, what’s the advice, or what have you learned, right, through, through navigating that?

Diane Tavenner: Well, I think that I mean, so much of what Dan and Cady have just said is so important. And I think that what, what was one key thing is, you know, I sort of set out to build Futre as an edtech partner that did things differently than what I experienced when I was sitting in, you know, the seat that Dan and Cady are in. And you know, that core value of our company is how we do the work is as important as the work that we do. And so how we do the work is very much co-building with schools and leaders and students. And so, you know, we are out in the field working with students and teachers and people like Dan and Cady literally every other week. So we are literally co-designing and code building what happens. And so what you just heard, that Futre is being designed to help young people build this identity over a 10-year journey. I mean, that’s unheard of, I think, in any sort of tech market.

People don’t think about that. We have real outcomes that people are aiming towards, and most tech products just look at what’s something that exists and try to make it more efficient or slightly better. They don’t think about the integration of it, the flexibility of it, how it will be used by the adults. I mean, As an example, they just told you Futre can be used both in individual coaching, mentoring, advising, counseling. It can also be used with groups of students in a classroom, and it’s actually literally designed to support both of those. And I will say the, the inclusion of really supporting real-world experiences came directly from our engagement with our school partners and our students. That emerged as this real need And we were watching people literally running around schools with laptops on their arm and all these spreadsheets and trying to organize. And so we have co-built these elements together.

But you’re right, the incentives in the business side of things are not to build this way. And so, you know, like always, we’re going to see if we can prove that wrong and say, no, when you do build this way, you not only get better outcomes for young people, schools and teachers and educators, but you also can be a successful, scalable product.

Michael Horn: So certainly a more enduring product if you, if you thread that needle, right? So for sure.

Cady Ching: Yeah, exactly. So I think it’s I think it also speaks to why it’s so important for Dan and I to sort of pull together a coalition of the willing with other operators. One thing we haven’t spent 鈥� I know we’re almost at time 鈥� that much time talking about is how hard this work is. It is challenging, and we have so much to learn. We are not perfect. We are learning every single day. We are constantly seeking out other school systems that have similar visions for education, and we’re trying to learn from them. We’re trying to get out onto their campuses and be in community with them because we know that if we want to build something that’s enduring and lasting and maximizing impact on the number of students in our country, or even globally, we have to build for the students of Summit as well as all students.

And I think that, that’s what’s most important for me as I set out to lead some of this work is if it only works at Summit, it’s not good enough. And what we’ve learned about leading change at scale is that we need a shared purpose for what school is actually for, and that belief that it’s possible to build a system for that purpose, which is actually no small feat. And it’s why we’re spending so much time building what I would call a coalition of the willing, which is educators and systems who agree on our common destination before we start building the actual tools. I think my core idea is that beliefs come first, model comes next, and then the tools come last. And when we get that order right, that’s when the scale can become possible.

Summit Learning: Model vs. Technology

Diane Tavenner: Cady, I want to double-click on what you’re saying because, you know, you talked at the top of this about how Summit Learning had really scaled across the country to 40 states and, you know, 100,000 students, etc. But Dan, you also said the technology, the Summit Learning platform was not the model. It is not the model. And the model has really taken root even as that particular piece of technology has gone away. That said, I do know that you both believe deeply that having an aligned core technology that is the infrastructure that sort of I think, Dan, you used the word guardrails, like puts up the guardrails and the support for the model is profound. And I know that you’re in conversation with other folks who’ve done some at learning who are, who it’s taken root for them as well, but are having a hard time really keeping that model intact. And so talk about sort of the need for that infrastructure, the role that it plays and what you think it might look like in 3.0. And Cady, you just said it, no one’s going to build technological infrastructure for a single school or a single school system.

And so there has to be this coalition.

Cady Ching: We have to create the market.

Diane Tavenner: Yeah. And so talk about that because the market generally is not very coherent. And as I sit on the other side, it can be really confusing and hard so talk about how you guys are thinking about that.

Enabling Learning Through AI

Dan Effland: Yeah, I think this is something we’ve started to be spending more and more of our time on as we’ve gotten clearer in the work with our students and caregivers and educators this fall. We’ve gotten clearer about where we’re going. There is this need, which is that technology is not the model, but it is, you know, there’s a reason we talk about time, talent, and technology as the big levers with resources. It is a huge enabler. And I think the possibilities with AI as part of that technology infrastructure make it an even stronger enabler. So I’ve already talked about like the idea of like dynamically reallocating resources, which is, I think, I love in a conversation educators here, because I think sometimes it’s not the, like the shiniest thing to talk about, but we know that getting kids the right thing at the right time in the right sequence is often the difference between learning and not learning, between progress and not progress, and between finding that pathway and not finding it. And so, at a high level, when we’re thinking about that infrastructure, we need to make sure that, like, we have a really rich, you know, amount of data.

And there’s a lot of work to be done there. Our school systems historically have not put data together in ways where you can create what like a technology person would call the data lake in a way where you can really access that as you need it. And then the next element is going to be a really robust knowledge graph that is not just academic standards. It’s got to be much broader than that. And then, of course, the way that AI would then interact with that to allocate and think about your resources. And I’ll share too, like when we think about resources, I generally think of everything as a resource. My time is a resource, Cady’s time is a resource, our educators’ time is a resource, curriculum is a resource, YouTube is a resource. Anything that can help a young person move towards those outcomes, we think of as a resource, and how can we constantly repackage those and get them in the right order while holding onto the vision? Because I think there’s a version of personalized learning that I would call like individualized learning.

That’s not what we’re talking about. I believe this has to happen deeply in community and with really strong relationships and human connection. And so the personalized learning, then it’s actually more complex when you’re committed to maintaining community and relationships, because you’ve got to figure out configurations of young people and not just put everybody separately on a computer they have a particular pathway and so.

Cady Ching: And that’s what we’re seeing, we’re seeing people just run, sprint towards an outcome without doing the diligence. And I think that it’s resulting in a lot of binary. If you’re either tech-forward or you’re human-centered, and there is a way to bring that together and build a model that’s doing both and that’s what we’re setting out to do.

Dan Effland: Yeah. There’s another binary too, that we haven’t talked about, but we should stamp here, which is this binary of like, real-world readiness or academic foundations. And that we now, we have these camps and like, we’re all about academics and we’re all about the real world. And when you talk to students, you talk to students and caregivers and educators, no one thinks it should be an either-or. That’s the scarcity mindset we’re often in, an area that we engage in educators. And we’re deeply committed that our young people will be prepared with college-ready academic foundations and real-world readiness, which means for us habits of success, communication, collaboration, all executive functioning. That is has a purpose

Diane Tavenner: Yeah. One is, as Dan, your story of that student showed, the sense of purpose, which is connected to what my life will look like in the future, really is what drives everything for a young person, right? It’s how they’re forming their identity as they build that vision. It’s what motivates them to stick to the hard work every single day on this journey to get where, where they’re going, and so yeah, I think what you’re up to is really critical. I hope that a lot of schools and systems engage with you to create this demand in the market for this type of infrastructure, dare we say, you know, Summit Learning Platform 3.0 as well. Because I think that it’s really, it’s hard to conceive of a post-AI model that doesn’t have that. That real infrastructure.

And I know you all haven’t seen it or found it yet, but continue to make strides in bringing it to life.

Michael Horn: This season of Class Disrupted is sponsored by Learner Studio, a nonprofit motivated by one question: what will young people need to be inspired and prepared to flourish in the age of AI as individuals, in careers and for civil thriving. Learner Studio is sponsoring this season on AI and education because in this critical moment, we need more than just hype. We need authentic conversations asking the right questions from a place of real curiosity and learning. You can learn more about Learner Studio’s mission and the innovators who inspire them at www.learnerstudio.com. 

So a good place maybe, Diane, to wrap up.

Should we pivot to our before we let you off the hook section? Cady, Dan, we have a tradition here where we, where we talk about something we’ve been reading, writing, watching, listening, whatever it is, not writing, listening to, and eventually I’ll get my verbs correct. But and then, so just often we try to keep it outside work, but we often fail. So, Cady, you want to go first, and then Dan, we want to hear what’s been on your playlist or bedside table, and then Diane and I will wrap it up.

Cady Ching: Yeah, sounds great. I have been鈥� I taught my 7-year-old what it means to brain rot. I don’t know if you’ve heard that term, but where you just sit on the couch and just kind of watch nothing for hours and hours. And we did do a Spider-Man and Avengers binge this past weekend. So that is something I have been watching a lot of. Reading is going to be hard for me to separate it from the professional. I’ve just been really deep in leader succession. I think to do this work, you need really strong talent in leadership pipeline.

And so I’ve been in HBR. I check the Marshall Memo every week to see what, what they’re pulling out, to really think about how I’m leading personally, locally, individually, but then also what the sector needs. Dan, I’ll pass it to you.

Dan Effland: Similarly, like the kind of first answer on my mind is just this fire hose of like white papers and podcasts about education and AI.

Cady Ching: And then he screenshots them and sends them to the whole team.

Dan Effland: Yeah, drive everyone nuts with them. But I do have a more, maybe a more fun one on the personal side. Kind of finally reading the Foundation series, the Isaac Asimov kind of classic sci-fi. It’s honestly about connection for me. My siblings are sci-fi readers and I’m very late to the party. And then my father is retired now, and one of his, it seems like, main activities as a retiree is to reread everything Asimov ever wrote multiple times.. And so for Christmas this year, I got a stack of these really great, Half Price Books paperbacks of all the Foundation novels, and I’m starting to work through them.

And we have a text thread about them, and they are, it’s a wonderful story, it’s very complex, and it certainly does also make me think a little bit about the future of our world and AI and, and what, you know, where, where young people fit in that, but it’s also just been a really fun way to connect to the family.

Michael Horn: That’s cool. Wow.

Diane Tavenner: What about you, Diane? Well, picking up on that. So first of all, apparently this is not going to be a novel recommendation because this Apple TV series, I guess, is the most watched at this point. But we watched Pluribus, which was created by Vince Gilligan, who 鈥� yes, Breaking Bad. Yes, Better Call Saul. I didn’t watch either of those, but I was a huge X-Files fan

Michael Horn: Back in the day.

Diane Tavenner: OK. And so there is very much some X-Files feel here in Pluribus. But to what Dan said, and I think Foundation is related, I just find this series to be so provocative in the questions that it’s bringing up and sort of the contemplation of where we’re going as a society and how the choices we’re making each day might affect that and what we actually want. And I will鈥� I told you I would report back my goal. I did finish Ian McEwan’s novel that I pre-promoted. Yeah, yeah, yeah. But it was everything I expected and more.

It was just extraordinary. And I did both of those over the holiday. And I will tell you, I feel like I’m sort of in surround sound right now of asking these big existential questions along with everything from what’s happening in the news on a day-to-day basis to all the work in AI. So, but I would highly recommend it. Super provocative and interesting.

Michael Horn: Perfect

Diane Tavenner: Perfect. Crazy. Like, you never know what’s gonna happen next.

Michael Horn: That’s fun when you can’t predict it coming.

Diane Tavenner: Yeah.

Michael Horn: Yeah. Yeah. I was gonna say, so the brain rot theme that you brought up, Cady, I mean, we talk about it all the time with our 11-year-olds, here at home. But I was 鈥� this is not where I was going to go at all with this, but I 鈥� something one of my kids said made me think of the Animaniacs theme song, if you all remember that cartoon from back in the day, and I pulled it up and showed it, and my wife just dismissively said, this was brain rot when we were growing up. so, there you go. the one I’ll say is, we all went with another family and saw Wonder, at the American Repertory Theater. Many people may know the book, Wonder, which follows the story of Auggie Pullman, a 10-year-old who has Tretcher Collins, syndrome that presents as disfiguration of the face and sort of how going into a school environment for the first time and all the things that it does. And there’s a movie about it as well, but now there is a musical too.

And Diane, you will not be surprised, I was crying from the opening number and I kept it up through the whole thing. So it was, I was true to form. That’s a good one to cry over. It was good. I represented well, but it was fantastic. We’ll see if it makes the jump from sort of off-off-Broadway to something bigger, but until then, if you’re in the Cambridge area, definitely check it out. And for all of you, just huge thanks, Cady, Dan, for joining us, getting us to have a peek under the cover of what’s coming next at Summit and the broader 鈥� as usual, you all are thinking about the broader ecosystem as well, which I admire so much about the work you all do at Summit. It’s not just our model, but how does our model spur this greater change across education.

So huge thanks for joining us. And for all of you listening, keep the questions, comments coming. Diane and I feed off them, and we really appreciate all of you. We’ll see you next time on Class Disrupted.

Disclosure: Diane Tavenner founded Summit Public Schools and served as its CEO from 2003 to 2023.听

This episode is sponsored by LearnerStudio.

Within weeks of ChatGPT鈥檚 release, hard rules appeared almost overnight. AI tools were banned throughout departments. Teachers watched what seemed like an existential threat materialize in real time, and they responded the way institutions usually do under pressure: They drew a line and told everyone not to cross it.

Three years later, that line is still there. And at many places, nobody ever asked whether it should be, at least not the people most affected by it.

When I looked into how my Austin, Texas, high school鈥檚 AI policy was developed, I found that my administrators made the decision internally. There was no student committee, no open forum, no campuswide survey. The rulebook was simply handed down. In K鈥�12 education, require districts to develop and publish AI policies; when they are published, they鈥檙e often developed without proper consideration of all stakeholders, including students themselves.

It鈥檚 reasonable to counter that students are minors, that institutions need coherent governance and that not all decisions can go to a committee. But AI policy isn鈥檛 a routine curriculum adjustment. It governs what tools students are allowed to use to think, draft, research and communicate 鈥� tools that increasingly shape how knowledge is produced and evaluated outside school. Getting those rules wrong produces consequences for students.

Brittany Carr鈥檚 situation is a well-known example. In early 2023, the had three assignments flagged by an AI detector. She provided her revision history and explained her process writing deeply personal essays about her cancer diagnosis, her depression and her personal recovery. It wasn鈥檛 enough. Fearing that a second accusation could cost her financial aid, she began running every essay through an AI detector herself, rewriting any sentence it marked until her writing voice felt flattened and unfamiliar. By the end of the semester, she left the university.

Carr is not alone. The same NBC News investigation found that students across the country deliberately simplified their vocabulary and avoided complex sentence patterns 鈥� not to write better, but to write less like themselves. Creative writing assignments exist to help students find their voice, which they can鈥檛 do in fear of an algorithm. Carr鈥檚 case shows a student reshaping her writing, and ultimately her education, around a software system she had no role in approving, in a policy she had no voice in developing.

Student involvement would not necessarily have guaranteed a different outcome in Carr鈥檚 case. But it might have changed the structure that enabled it. Students could have brought up concerns about relying on automated detectors without corroborating evidence. They could have described how fear of false accusations pushes students toward simpler vocabulary, safer syntax and less intellectual risk. They could have asked what procedural protections exist before a software flag becomes an academic charge.

Instead, at many institutions, enforcement architecture was built first. Conversation came later, if at all.

It doesn鈥檛 have to work this way. In Los Altos, California, did more than sit in on policy meetings 鈥� they designed and ran community workshops, facilitated discussions between sixth graders and administrators, and built an AI chatbot to help other districts draft policies. 

A found that students overwhelmingly want to be part of decisions about how AI is used in their education 鈥� and that many already hold sophisticated views on its risks and potential. The fact that Los Altos made national news tells you how rarely that invitation is extended.

But there is a deeper reason students belong in these conversations: We know something policymakers don鈥檛.

At my high school, I鈥檝e witnessed 鈥� and experienced 鈥� a secret loop in the learning process: we use  large language model tools like ChatGPT and Claude to genuinely improve learning by unraveling concepts, studying for tests and brainstorming ideas. 

A few days ago, a student asked a question about a formula in my AP Physics C class 鈥� and nobody knew the answer. Another student opened his laptop and asked Claude, and after a few minutes of back-and-forth, we had completely straightened out our question, improving everyone鈥檚 understanding of how circuits worked. I used an LLM to compile notes from my Multivariable Calculus class, which helped me study and earn a near-perfect score on my test. My friend used ChatGPT to learn Java syntax for a project 鈥� not to write code, but to understand the language.

A found that 54% of U.S. teens now use AI chatbots for schoolwork, with the most common uses being research and brainstorming 鈥� not copying and pasting answers. But that message hasn鈥檛 reached the people writing the rules. This secret loop goes completely disregarded by schools, simply because it鈥檚 easier to blanket-ban the technology altogether. The generation that grew up with these tools understands their texture in a way no outside committee can replicate.

These AI policies directly affect students鈥� outcomes and futures. To exclude them from the conversation is simply undemocratic.

If educational institutions are serious about preparing students for democratic citizenship, that commitment must go beyond coursework and into policy-making. The time to invite students into these critical conversations is now. Will schools treat students as subjects of policy, or as participants in it?

We are developmental cognitive neuroscientists who study how children鈥檚 daily interactions with parents, caregivers, teachers and peers shape , and development. We are not anti-AI, but we are extremely concerned about corporate efforts to market AI toys to parents and educators of young children. We do not yet know how many young children are already engaging with generative AI bots, but if are any indicator, this is a rapidly growing market. 

Some companies say their toys and devices are 鈥渁ge-appropriate鈥� and will support children鈥檚 learning and development, but that鈥檚 not always the case. For instance, the makers of Kumma, a plush teddy bear, promised to build conversational skills for children from ages 3 to 5. But the toy was pulled from the market last year after it was caught encouraging researchers testing it . 

Beyond these physical safety risks, we have essentially no data on how interacting with generative AI 鈥渇riends鈥� will shape very young children鈥檚 foundational brain, socioemotional and language development. Rather, the preponderance of evidence about how brain development works in the earliest years of life suggests that families should proceed with caution before letting their littlest children play with these new technologies in the form of toys.

We are not alone in this concern. Together with scientists around the world who study the exquisite, human-to-human interactions that shape early brain and cognitive development, we recently released an about the risks of direct infant-AI interaction. 

Decades of scientific studies paint a clear picture of optimal development in the first few years of life. Babies and toddlers grow and learn through daily, moment-to-moment interactions with their close caregivers. Indeed, humans cannot develop fully without these foundational interactions. Present, responsive, real-time interactions shape children鈥檚 language, sculpting their growing understanding of new words, grammar, pronunciation and social intentions. 

These real-time interactions shape children emotionally, helping them map their inner experiences to their outer perceptions. There is evidence that when a caregiver and a young child interact, 鈥� from eye contact to to heart rates, oxytocin levels, and even . 

Unlike AI models, which can parrot human-to-human interactions, caregivers pair their words with touch, eye contact and facial expressions that signal their love and attention. Real conversations include inside jokes, local dialects, family lore, and the distinct conversational patterns that make a family a family and a community a community. 

Development is about real-time rhythm, and every unique caregiver-child dyad develops their own. It鈥檚 not about perfection. It鈥檚 about presence, something an AI model can never and will never be able to provide. 

In fact, toys that imitate social responsiveness may interfere with an infant鈥檚 developing sense of how people relate to one another. The better these toys get at mimicking a parent, a child care provider, a grandparent or other adult caregiver, the more concerned we should be, particularly in the earliest years when infants and toddlers are developing a distinction between self and other  鈥� a growing awareness that the other humans who surround them each have inner worlds of their own. 

From a policy perspective, . There is much more to learn about these new technologies before parents let their babies play with them. 

Without these policy protections, parents and educators must take the lead, that simulate social reciprocity, replace face-to-face caregiving, or are designed to replace soothing behaviors that infants and toddlers need from caregivers in order to build attachment, trust and human connection.

The earliest recorded scientific experiments with electricity happened 3,000 years ago. Today, access to electricity has raised the standard of living for nearly the entire world. Still 鈥� after more than a hundred years of widespread use, safety standards and engineering to wield electricity for the common good 鈥� no responsible adult would let a child anywhere near it in raw form. 

AI has the power to improve human lives, but these are early days. We take for granted that we cover our light sockets to protect all our community鈥檚 children. We must take the same protective stance with AI.

New York City鈥檚 Education Department unveiled its for artificial intelligence use, offering a rough road map for if and when to incorporate AI tools in school.

The guidance, released Tuesday, arrives nearly three years after a short-lived on ChatGPT. It also comes in the midst of ongoing debates about student privacy, AI鈥檚 effect on student learning and development, and the role of private companies in schools. Some schools had as they awaited citywide guidance.

Hot button issues, like how and if students can use AI for homework assignments, or whether students can use personal AI chatbot accounts in addition to tools approved and supervised by the Education Department, are still being hashed out.

City officials are asking families and educators for feedback, which will inform future versions of the guidance. The Education Department released a and will also host webinars and events to answer questions and gather feedback through May 8.

鈥淎I is here, and our responsibility is to put strong systemwide safeguards in place,鈥� schools Chancellor Kamar Samuels wrote in an email to parents.

The early framework is structured in a 鈥渢raffic light鈥� approach: green light for approved uses, red light for prohibited cases, and yellow light cases for gray areas, which require significant oversight.

For example, brainstorming lesson plans and drafting non-critical communications fall under 鈥済reen light鈥� cases.

In 鈥測ellow light鈥� cases, schools can use AI to find trends in student data, to generate translations for bilingual learners, or adapt materials for students with disabilities 鈥� but a trained professional must first review the outputs before it is used with students.

All decisions made about students, including grading, development of special education and 504 plans, discipline, counseling and crisis intervention, and other academic placement decisions, are strictly forbidden. These 鈥渞ed light鈥� cases are not expected to change in the final playbook the city aims to release in June.

Pushback has already been fierce among parents and education advocacy groups: A asking the city to put a two-year pause on AI use in schools has garnered about 1,500 signatures since October. Several Community Education Councils have also passed resolutions calling for a moratorium of AI in schools.

The guidance was written by the Education Department鈥檚 AI Task Force, and informed by the city鈥檚 external AI Advisory Council, which includes education technology partners from Google, OpenAI, and other companies hoping to contract with the city鈥檚 roughly 800,000 K- 12 students.

Questions remain about student privacy and third-party AI contracts

Before schools can use AI tools in the classroom, each product must go through a data privacy and security vetting process called the Enterprise Request Management Application. The process, created in 2023, applies to all third-party technology vendors.

But AI has become ubiquitous. The Education Department鈥檚 contract with Microsoft 365 programs did not originally include AI chatbots, but now do, said Naveed Hasan, a member of the Education Department鈥檚 Data Privacy Working Group.

鈥淛ust like TikTok was unregulated until school networks blocked it, so are these free AI products,鈥� said Hasan, whose group advised on data privacy policies prior to the AI guidance.

Schools can visit the department鈥檚 to see if a tool has already been approved; otherwise, schools must submit an application for new use.

The process, however, doesn鈥檛 yet include guidelines on how to review certain aspects of AI products, such as algorithmic bias or instructional effectiveness. Those are expected to be included in the final June version of the playbook.

The guidelines, which were shaped by federal and local laws, say personal student information can never be entered into unapproved AI tools, and under no circumstances can student information be used to make money or train AI models.

Although the general sentiment about privacy protection is clear, how to ensure it remains protected in every use is a key question that some close to the policy development say remains unfinished.

Hasan said the guidance alone can鈥檛 guarantee privacy and relying on third-party products, even approved ones, makes it difficult to know what鈥檚 secure and what鈥檚 not.

He has called on the Education Department to consider maintaining its own hardware and training its own group of AI experts instead of relying on outside companies.

AI moratorium advocates push back

The Parent Coalition for Student Privacy, one of the groups on the AI moratorium committee, said in Tuesday that the guidance does not address the potential long-term effects of AI use on learning and thinking.

The city has already accepted that AI will be a part of school learning before proving its value and safety for students, said Kelly Clancy, founder of Parents for AI Caution, another group on the committee.

鈥淭he city needs to have a burden of proof about why this is good,鈥� Clancy said. 鈥淚t shouldn鈥檛 just be about harm reduction, but rather why AI is better for my kids than a human-centered, traditional classroom.鈥�

Education Department officials said proposals for new, AI-focused schools and programs 鈥� like Next Generation Technology, an 鈥淎I-focused鈥� high school 鈥� must demonstrate how they align with the guidance鈥檚 principles.

The full preliminary guidance can be accessed .

Chalkbeat is a nonprofit news site covering educational change in public schools.

John Danner, the cofounder of Rocketship Public Schools and now the founder of Flourish Schools, an emerging network of AI-native microschools, joined Michael Horn and Diane Tavenner to share what鈥檚 now possible when it comes to school design in the age of artificial intelligence that wasn鈥檛 previously possible. Danner explained how Flourish is leveraging AI to deliver foundational skills like reading and math through conversational tutors to free up teachers to focus on building relationships and nurturing students’ passions and “superpowers.鈥� 

He also shared how they鈥檙e using the technology to provide real-time assessment and feedback on student projects. The conversational models can be much more powerful, he says, than previous edtech applications. 

Listen to the episode below. A full transcript follows.

Diane Tavenner: Hey, Michael.

Michael Horn: Hey, Diane. It is good to see you again for our continuing conversations on AI.

Diane Tavenner: You too. This one’s going to be a fun one. You know, our most recent episode, we talked with Alpha School founder Mackenzie Price. Most people have heard of Alpha at this point. It’s getting a ton of attention. And so what we tried to do there was really move beyond the talking points and the marketing to really dig into the model itself, including specifically how they’re using AI, which is turning into a bit of our quest this season. And so this conversation today is a part of that exploration on who’s building what I would call maybe AI-native school models, if anyone. And, you know, what might they look like? What are they starting to look like? And it’s a really fun conversation today because we get to have a chat with an old friend.

Michael Horn: Yes, that is indeed correct, Diane. Today we’re going to get to chat with none other than John Danner. John, for those that don’t know him, has had a decorated career in tech before turning to education, as he co-founded and led NetGravity, the first ad server company, I believe. And after taking it public, selling it to DoubleClick, John went back to school and then became a teacher, and he taught in Nashville for a few years there. And then I think a lot of folks know him because he co-founded, of course, Rocketship Public Schools in 2006, which we, of course, talked about also in our last episode. But Rocketship was a buzzy school for a good while there, marked by its student outcomes, its use of technology, its expansion. And then after leaving Rocketship in 2013, John did a number of other things, including founding an online math tutoring company, creating some very interesting education investment vehicles and more. But I want to skip ahead to his most recent venture, Flourish Schools, which is what we’re going to hear about today.

Michael Horn: So, John, hopefully I did some justice to the bio, but, welcome. It is always good to see you.

John Danner: Thank you, Michael. Great to see both of you. Long time.

Michael Horn: This is going to be fun. This is going to be fun. So let’s start with grounding our audience. My assumption is that a lot of folks know Rocketship and what you did there. Far fewer know about the Flourish Schools model itself and what these schools actually look like. So maybe give us the basics, like what is Flourish Schools, how many of them are there today, how big are they, what’s the grade levels, what does a day in a student’s life look like at these schools? You know, paint the picture for us.

John Danner: Yeah, yeah. So we started Flourish about a year ago. We opened our first school last August. In Nashville, one microschool so far. They’re middle schools, so grades 6 through 8. I’m out in Phoenix today. We’re opening a couple more schools in Phoenix next year, next August. And I’d say the reason for doing it, you know, Diane knows this well, like doing schools is quite difficult work.

Enhancing Foundational Learning with AI 

John Danner: I often prefer being on the software side where, you know, life is good. But, you know, schools are hard work and sometimes you have to do them. I think the big motivator in starting Flourish for me was that I had started a couple of AI companies, Project Read, probably the most notable doing reading, which is in a lot of classrooms. And I just noticed that most schools are using AI in a very supplemental way right now, very much the same way they used edtech. And that bothered me because, you know, in reading, for example, I think there’s a pretty good argument that AI for reading is going to be better than the best human reading teacher within the next year or two. It’s not a long way off at all because teaching reading is really hard. Training teachers to teach that is hard. It’s hard to be patient with kids when they’re making lots of mistakes.

And it’s hard to remember everything a kid has ever done when they’re reading with you, right? All of which just is default for AI. So, you know, in watching Project Read roll out and seeing everybody kind of use it, you know, in those last 15 minutes in the class when they were kind of, you know, a kid was done with the assignment and needed to do something else. Like, I was like, you know, that doesn’t seem like how AI should, affects schools. It should be used more strategically. You know, what can AI do, and therefore what do you do with teacher time? I think, you know, for me, teacher time has always been kind of the scarce resource. It’s like whatever teachers focus on is really what schools do. No matter what schools talk about, it’s like, OK, what, what are your teachers doing? That’s what’s going to have the most impact. And so Flourish we, we started with the assumption that what we call foundations, kind of the basic skills, reading, writing, math, are going to be better taught by AI.

The way we kind of look at it is if you think of like Tier 1, Tier 2, Tier 3 instruction, it’s really the move from technology as a Tier 2 or Tier 3 product to a Tier 1. So, you know, can you use AI to do kind of tier 1 basic skills and standards-based instruction? And so that was what we did from day 1 at Flourish. We’re 6 months into it now. I would say the lesson learned is, of course, you’re going to have students in any school that like, you know, whatever. We have several special ed, several ELL students they need more time and attention. But during our foundations block, which is an hour long, teachers have time to work with them one-on-one. And a teacher working with a student one-on-one on reading or whatever is like a luxury that like no other school has because that you can’t have them doing that. But when all the other kids are making great progress with AI, having a teacher spend that time, that luxurious time is actually possible.

AI’s Impact on Schooling

John Danner: So that’s the fundamental thesis is that we can do that in a way that that’s what our teachers are not doing and spending all their time preparing for and teaching during the day. And that allows us to kind of come up with a new curriculum. And I think actually, you know, you guys want to focus on AI and we should. I think the actual interesting question with schools is once you make the commitment that AI is going to do a lot of this basic instruction, then you’re confronted with the now what problem, which is like, oh gosh, what’s school for like moving forward? And I guess that’s, that’s what we’re kind of excited about is we’re in this super serious time of change for students. They’re not going to grow up to a world that we all experienced. You know, my daughter just got out of college. She was a pre-med, but didn’t really want to be a doctor. She gets out in the job market and gosh, there are no jobs.

And like all those other things that she learned along the way about hustle and, you know, you got to go put yourself out there and whatever played out and she found a job. But boy, like if you had just spent all your time in school, like learning algebra or whatever, she wouldn’t have done well. So, I think, you know, our point of view at Flourish is we, we talk about 3 things mainly, relationships. So these are middle schoolers. So how do you get along with other people? And we do an hour we call circles, which is really as kind of therapeutic as it might sound, where kids are sitting in a circle talking about their feelings, how other kids affect them, et cetera. And for many, many of our students, I’d say it’s pretty mind-blowing to actually understand how other people are thinking, you know, as you’re talking and saying things and stuff like that. Really powerful.

So relationships are a big piece. And then we talk about two others, superpowers and passions. So superpowers is kind of our word for what people have called soft skills. I hate the term soft skills because it’s kind of denigrating in a world of like standards-based instruction. Oh, that’s the other stuff that, you know, makes you a human, but it’s not nearly as important as high school chemistry or whatever. Like, we actually think it’s the opposite now that knowledge is pretty abundant and accessible, like the things that make you human are the more important things. So, do you have agency and curiosity and these other things that make you awesome? That’s important. And then the passion side is really, what do you want to do when you grow up? What are you excited about? What are your big interests? Which, you know, as you know, for upper-income families tends to happen at home.

You know, you’re sitting around the table or you go, you know, on a little family field trip or whatever, and kids are discovering lots of different things that they might be excited about. Happens a lot less in working class and lower income families. We’re purposefully mixed income. We took a page out of your book for that, Diane. I think that’s really the right way to do this. And so for our kids who are, you know, working class and lower income, we think like discovering, what the world is and what you might want to be in is super important, especially in middle, so that you kind of enter high school with some idea of like what you’re excited about and some kind of path you might want to pursue. Even if that changes, that’s OK, you’re not just kind of clueless showing up in high school, which, you know, a lot of kids are.

Diane Tavenner: Yeah, super helpful, John. You know, one of the ways I’ve been trying to have conversations with people about what these sort of AI-native models will look like or can look like or do look like is I don’t want to have a conversation where we compare what they’re doing compared to like the old industrial model classroom, right, that’s like not useful to me.

John Danner: We’ve had that conversation. Yeah.

Diane Tavenner: So I keep using the sort of Rocketship and Summit because I know them the best of like best-in-class sort of personalized learning models that we were doing the very best we could at the time with the resources we had, and doing a lot of what you just described, right? Like, I’m assuming circles maybe comes out of Valor, which, you know, it has, you know. So like, a lot of that great stuff we were doing before. So what I’m really, and you’ve alluded to this, I think, with shifting Tier 1 instruction out of the classroom model and the AI is doing that. But let’s dig in a little bit deeper. Like, literally, what’s possible today that we just didn’t do 10 years ago and now we can do it? And what does that specifically look like in the model?

John Danner: I think the big change here is really one from point and click to conversational, right? Like, that was the eye-opener for me, really, you know, back in the ChatGPT moment was you kind of just immediately it became clear that a conversational agent would be able to kind of work through things with a student in so much better way than, you know, kind of what we all did with kind of edtech back in the day. So, you know, we all, we call it personalization, but there’s kind of a difference between a program more or less knowing where you are and what you need versus what an AI does, which is it knows everything. You know, like in Flourish, we more or less pour everything about a student into it. We have transcripts from everything students say. Like, the AI just is all-knowing about what’s happened with that student at the school. And so when it’s personalizing, it’s 100 or 1,000 times deeper level than like this basic categorization that edtech used to be able to do. So I think it’s much more aware of what students need. And I just think the mechanism of talking to a student conversationally is so much better than kind of navigating through a bunch of screens and the stuff we used to do.

Diane Tavenner: So I’m assuming then you’re building your own. It sounds like you’re building, you called it curriculum, but like that tier 1, because I have yet to see sort of off-the-shelf products that are really, that I would be like, yeah, they’re great. They can do the tier 1 instruction. Talk about what you’re building, what that looks like for middle school kids, you know.

John Danner: Yeah, right. And remember, we’re 6 months old, so anything I tell you is like total work in progress. But, you know, we’ve got good people and we’re working pretty hard on it. So the, you know, the fundamental idea, so I’ll tell you where we started with this and then kind of where we are now. We kind of had this idea that we’d have an agent on our side that was very good at sending kids to the right place to get the right help, right? So kind of like a hybrid between the old ed tech world and kind of this AI-driven world. And we pretty quickly discovered the kind of things that we had discovered at Rocketship, or I’m sure you did at Summit, which is there’s so much friction and stuff involved in manipulating another program. It’s like basically not worth it. And so that probably took a couple months for us to just realize like this is a waste of time.

Tutoring via Adaptive Dialogue

John Danner: And so really the way our system works today is as a student, I’ll tell you today and then where we hope to be in 2 months. So today, the way it works is that we have kind of a pre-assessment where we’re looking for what a student knows. Based on what they know, they enter a conversation with our AI. We often will have a 1 or 2 minute video of like just what that thing is, kind of an old edtech type thing, right? Just because I think a framing is often helpful for a new concept, but that the majority of the real instruction is kind of this dialogue between the AI and the student on like, OK, well, let’s talk about, you know, two-digit addition just for lack of anything better. Here’s a problem, you know, solve this problem for me, tell me how you’re doing it. And then basically just digging in as the student doesn’t get it. And it’s so easy to prompt for, I mean, you know, Zeal, my third company, the math tutoring company, we had figured out all the misconceptions that every student has in math. And so when you prompt an AI with that, OK, here are the 10 likely things that a student’s going to do wrong, when they’re doing two-digit math, it just goes, oh, OK, that’s it, and then it goes deep there, right? So if you think about it, it’s very fluid.

It’s very much what a human tutor would do in that case. They’re kind of responding in real time to what that student’s doing and going, oh geez, you don’t really understand how to carry the tens place, so let’s go deeper there or whatever. So that interaction with the AI happens, and then we go out and post-assess. And so the student’s kind of manipulating where they want to go and what they want to do through that process. Where we’re going, where I hope to be in a couple months, is that that’s all, all the pre- and post-assessment is kind of gone. We’re finding that the AI through that dialogue has just as good an understanding of what that student is capable of doing as kind of any formal assessment process. And it’s much more natural to just have the students sit down with the AI, you know, when they start and talk about what they want to work on. And then, you know, kind of the AI drills into that and shows them a video and does things like that.

So I think it could feel quite a bit like, you know, a student showing up at a tutoring center and that tutor kind of just working with them. It feels like that’s going to work. But that’s where we’re at with it.

Diane Tavenner: Is that voice or are they typing or both?

John Danner: We’re doing typing now. We’d love to do voice. We started there and we really worked hard on it. I would say that the biggest problem with voice for us is that we have never figured out the kind of noisy classroom problem. Very hopeful that somebody does because of the issue, you know, even if you’re off in a corner of a classroom or even outside in the hallway, the AI hears everything. And so it you know, and if you think about it, like when you’re in one of these sessions, the AI hears something and somehow inserts that in the conversation. That’s just weird. It kind of ruins the whole flow.

So it’s easier with middle schoolers to do kind of a text-based one right now. But I, you know, what I’ve told the team is I think the main interface for AI will probably be audio at some point. Like it’s just the most natural way. And so as the industry kind of builds better and better models for that, I hope that this problem gets solved and we can go to audio.

Diane Tavenner: That makes sense to me. And do you then have a knowledge graph underneath that? So even though the students sort of like flowing where it makes sense to them, at the end of the day, you have kind of the macro plan of where you want them to go.

John Danner: And yeah, so we built a super elaborate one for Zeal and unfortunately are more or less rebuilding it now for all of our stuff. Yeah, I think that’s right. I mean, as you guys know, the real challenge with AI is often that it’s so good in the moment at these things, but you kind of have to bring it back to reality sometimes. And so, you know, having a prompt that says, hey, pull the knowledge graph and see what’s the most important thing to work on is helpful. It’s kind of like this, you know, savant type tutor that can help a kid in the moment with anything, but kind of loses the picture of like what’s the most important thing to do. So you kind of have to bring it back.

And I think the knowledge is the way to do that.

Diane Tavenner: John, how does this connect with, I know you’re very committed to project-based learning and sort of that approach, which you know that I am as well. And, you know, it sounds a little bit like what you’re describing. You know, at Summit Learning, we have the playlists where you were doing the content knowledge. What you’re describing, I think, is a stronger version of that and what AI can do. How are you connecting it to the projects? What’s the intersection there? What’s going on there? And are you using AI in the projects?

John Danner: Yeah, the answer to the second is definitely yes. And let’s talk about that in a second. So we have a theory as a, as a school system, that’s probably the opposite, at least the opposite of like my alma mater. I’ve been talking to Bellarmine. It’s my alma mater in San Jose, talking to teachers about that. And, you know, AI is a problem for a lot of schools and teachers, right? Like it’s the cheating and stuff like that. We have basically the opposite approach, which is like, assume any kid can use anything that will help them read, write, understand, research better, and then like uplevel what you’re teaching so that you assume that yes, everybody’s writing is going to be perfect now. Don’t worry about that.

That’s not your job anymore. So with projects, you know, the link really is when you’re in a project, you’re trying to apply knowledge to build something to do something. And it’s extremely common to not understand something well enough to do that well. And so you need to go off and kind of research and understand it. So the link that will exist that doesn’t exist yet, which I’d like to see, is foundations lives in its own block right now at Flourish, but we’d like foundations to be accessible kind of basically all the time for students so that that’s the main way that you research as well through kind of an AI interface. So that’s the ideal. Right now what happens is that a student kind of struggles, they go off and use Gemini or something for things. And then we know, you know, the AI knows because it’s paying attention to the project and what’s going on.

‘Oh, this student struggled with this,’ and then in Foundation that kind of bubbles to the top the next day. But like, why wait? Like, just make it real time. If a student’s struggling with something, just go ahead and do it. We do have to figure out kind of the, you know, the tier 1 versus tier 2 of this. Like, if a student’s really struggling and they’ve got a real issue and you just wipe out project time doing that, that doesn’t feel right either. So we’re gonna have to figure out like what level of intervention happens if, you know, they’re still not getting it. But certainly at least the tier 1, like, oh, I just don’t know about this, let’s learn more, should happen through that Foundation system, we think.

Diane Tavenner: That makes sense. Yeah, that makes sense to me. Tell me about what the educator is doing in these times.

John Danner: Yeah, I mean, I think that’s the most important thing really is And I know for many, many teachers, the concern is, gosh, well, maybe you just don’t need me anymore or something. And that’s just completely not true. I mean, I noticed this at Rocketship, you know, people go into teaching because they love kids. That’s like, you know, that’s the common thing that you always hear. Some people go into teaching because they want to be content experts, but not that many, at least at kind of elementary and middle, like, it’s still really driven by like, I really wanna connect with kids and be with kids, not like I wanna be the best reading teacher or whatever. And so, you know, when you kind of push a lot of this like content knowledge and instruction to AI, what really happens is a little bit of like what I was describing with tier 2 and tier 3 during that time where a teacher now has a lot of time. So, you know, a lot of the stuff is going on. Project-based learning is nice that way.

Building Teacher-Student Connections

John Danner: Kids are working on things, which feels kind of like a big Montessori classroom or whatever, where like everybody’s being industrious and getting things done. But like, you know, the question is always, OK, so like what’s the best and highest use for the teacher at that point? So I think, you know, our opinion in general is kind of building trusted relationships is the most important thing you can do as a teacher, right? Like anytime you think about teachers that affected you, it’s because for whatever reason they spent the extra time to kind of get to know you, understand what you were going through, and like became kind of a trusted friend and advisor. And I think buying time back to allow teachers to do more of that is by far the highest value. Of course, interventions and things like that are awesome. Having students reach to do higher-order thinking once they’ve finished a project, all that’s great, but I think it’s all in kind of service of making that connection between our teacher and our students such that the student is more excited and interested to, you know, learn and think with that teacher about other things, you know, especially with superpowers and passions and things like that. Like, we have it, I’ll just brief aside, you know, we have these report cards that have superpowers on them. And so they say things like, you know, organization or self-awareness or whatever. So you can imagine our parent-teacher conferences are pretty amazing because while a parent is like, yeah, I don’t really know much about middle school math and frankly don’t care that much.

Boy, when you bring up self-awareness or something like that, they can go on for a long time. And so you have these really deep discussions about these kinds of things and kids by middle school, certainly in high school, they’re not really listening to their parents about these things very much. They’re kind of sick of hearing this. So I really do think schools have a way better chance of kind of influencing how children are doing these things, especially around superpowers and passions. But that requires trust and trust, you know, it’s hard to build. So we think that the best thing for teachers to be doing is kind of like getting into deeper conversations with students and talking to them about like, you know, what their interests are, what they like. And building that in the hope that they have influence over that student’s trajectory.

Michael Horn: Well, so, John, I think this actually is perfect translation into the other thing that AI is doing to free up teacher time for that, which is, as I understand it, at least from, from what you’ve written, is that you have this AI coach that is quite involved in the project-based learning piece of this equation. And I think two distinct ways. So, maybe talk about that.

John Danner: Yeah, I mean, again, work in progress, so I’m not super happy with how it’s being involved right now, but I’ll tell you what I want it to be doing well. So I think that, you know, and Diane, you live this, that the real challenge with project-based learning is there’s kind of like this huge amount of really mechanical stuff that happens in project-based learning, whereas students are confused about what they’re doing, or they’re tired and not motivated, or whatever, and you watch project-based classrooms and like actually like 80% of the teacher time is like walking around doing that stuff where they’re like, come on, Joey, let’s get going, you know, blah, blah, blah. Which of course there will still be some of that, but to what extent can you create a really awesome thought partner that kind of does a lot of those things? Like, hey, Joey, you know, what we need to focus on here is this. Have you thought about, like, you know, kind of re-engaging the way a good teacher does. Because if you can free them of a bunch of that kind of, you know, really mechanical time, I think not only does it free time, it also like kind of frees your mind up as a teacher to kind of think deeper and like look for relationships and, you know, these kind of things that we really want teachers to do. So I think that’s a big piece of what we’re hoping that this coach does. The other thing it really does for us, and you asked about this before as well, Diane, is it listens. So we’ve got mics all over the place, students are talking, it’s all anonymized, but basically the system knows what bucket to throw all the comments that students are making, etc.

Teaching Soft Skills

John Danner: And when you think about like superpowers, these soft skills. One of the other difficult things in that kind of curriculum and approach is like, and you see it in kind of SEL-type schools all the time, it kind of devolves into like playtime sometimes where it’s not as rigorous. And what AI can really do there is by looking for evidence of, you know, perseverance, for example, when did the student show that they didn’t just stop, they kind of asked the next question and kept going? Like when the AI can provide those examples in each student’s kind of superpowers report card of those things and the teacher can review it, that is so helpful because, you know, when it comes to like pushing for students to improve in these areas. Teachers really have to know, like, kind of where everybody is, where is John on these different skills, where should I focus. And so helping to provide data so that teachers can do that is, is really, really important. I would say it’s pretty good. Like, here’s one thing that kind of surprised me, we did this like a month and a half ago, the AI assessing these, we have 24 of these superpowers across all the students in the school. And we did the AI-rated students on a scale of 1 to 5, and then 3 teachers rated those same students.

And it was only off from kind of the lead teacher by about 10%. So like you know, that to me, that’s like, it’s close enough. It’s kind of like stuff where it’s like, you’re probably right, like a super expert teacher can absolutely do a little bit better. But like, we kind of want to get it to the point where the teacher’s like, yeah, you know, I pretty much trust this. I’ll look at the evidence, but more or less, it says that, OK, like, what should I do about that?

Diane Tavenner: And John, that assessment from the AI was just sort of that natural capture of all they’re doing and assessing based on, yeah, to me, like, then assessment is a no-brainer. That should, I think it’s a conflict of interest for teachers to be assessing, quite frankly, but that’s another conversation. But,.

John Danner: I mean, the other point here, right, is that when you do assessment that way, I think it’s both more valid and stops taking classroom time, right? It just happens naturally. And that’s how it happens in the real world too. It’s not like you sit down and.

Michael Horn: You go, right, we don’t stop and say, now here’s your time.

John Danner: You don’t give somebody a 5-question assessment. 6 months or so. It’s crazy.

Diane Tavenner: Yeah, yeah. So, can I just play back to you what I think you’re just, saying, just to make sure I’m getting a real picture of what’s happening or what you are moving towards happening? And you’ve only been at it for 6 months, but you’re making pretty quick progress, it sounds like. So this, like, if I’m a student in my project time, and we all know this happens a lot, there’s some kids who, like, literally, you know, the teacher’s bumblebeeing around, and every time the teacher bumblebees around, maybe I’m productive for that moment, but then the teacher bumblebees away, and then I’m kind of playing or I’m whatever. But AI knows what I’m doing in those in-between times, and so I’m getting some sort of feed or feedback of some sort, and the teacher’s seeing it, my family’s maybe seeing it, of like, hey, this is what’s going on in your time, and so we’re going to hold the mirror up, give you some feedback, tell you like, this is the stuff you could be doing to be more productive. Is that kind of what you’re describing? And If so,

John Danner: Yeah, we’re all going to have that. So this is another thing, like one of the things we think about a lot at Flourish is like, is this different than the real world’s going to be or the same? And I think we all basically need that. Like, you know, if you had a voice that was kind of going like, John, what are you doing? You’ve been doom scrolling. You know, like it’d be pretty helpful, really.

Diane Tavenner: Well, one of the big conversations is about motivation, right? And like, oh, you can’t, you have to like motivate kids to use the technology to learn. But actually, I think you’re flipping the script here and saying like, no, the technology is like literally helping, young people be motivated because someone’s paying attention and they’re noticing what they’re doing and they’re giving them feedback on it. And you know,

Feedback and Rewards Drive Success

John Danner: The feedback thing is the important thing. It’s like basically if something’s giving you feedback, like even if the feedback’s not perfect, it’s so much better than not getting feedback. You know, like the classroom where everybody’s got their hand up and they’re just waiting for the teacher to call. Like that’s a bad place to be. So now you’ve basically got this continuous loop. The other thing I would say that I think is just almost for free in this world is, you know, the gaming world has figured out a lot of things that they do when you’re doing a pretty basic task to play the game, and you might not be that excited about it, but like, you know, they’re setting up rewards. We use badges, um, you know, so like an example is you might do 2 or 3 different projects, and by doing those 2 or 3 different projects that was built up to a badge. And so the badge is kind of hanging out there and some other student in the class got it.

And so you want it and things like that. And, and those like really kind of basic game things are very helpful at different times during the day, right? Like we kind of all need a little bit of push. We’re very conscious of intrinsic versus extrinsic. motivation. And so like projects are a good example where the default is intrinsic. We want students to be kind of working on that project because they’re interested in that, because they want to do it. But there are definitely times where the AI is paying attention and kind of prompting and even, you know, doing some rewarding and things like that is actually quite helpful for them to kind of persevere.

Diane Tavenner: John, I want to talk to you about, I think you’re the perfect person to talk to about this. So one of the things I hear out there a lot is like, oh, the hyperscalers are just going to build this. Like, number one. Number two, most schools and school systems have zero ability to actually build what you’re building. So you’re sort of this unique person because you sit at the intersection of like opening, operating schools and the ability to build sophisticated technology. Is that, are the hyperscalers going to build what you’re building? Like, are you, like, how do you think about the building of the technology here for schools?

John Danner: Yeah, I mean, we’d be pretty happy if the hyperscalers built it, first of all. We’re, you know, so I think that the main challenge over the next 20 years in education is going to be how quickly do we move to a world where students are living in the current world as opposed to the, you know, 20 years ago or whatever. Like, and, and so these basic things we’re doing like foundations, I think it’s important for students to live in that world now. And so what does it take school systems to move towards that world? I know that your approach at Summit, our approach at Rocketship in the beginnings of the edtech world were, hey, let’s just build these kind of basic model schools and hopefully people will come visit and go, oh gosh, you know, that doesn’t look too bad. Like I could probably do that as well. So I think a lot of the point of Flourish is creating this proof point where people can come and see and go, huh, that, that actually works well, and it’s definitely not dehumanizing. I see the teacher interactions with the students as being more human, um, than my classroom. So I think that’s like actually our point, our reason for being is to kind of be that model.

And, you know, we’ll build a network and we’ll get as big as we can, but, but really kind of purposefully influencing school leaders, district leaders, state leaders to think about, like, you know, what they could do as well. On the technology side, I’m generally of the opinion that a lot of this will get easier and easier for everybody who’s not at the foundation level over time. I will say, like, there are some exceptions to that. So, like, with Project Read, with phonemes and graphemes. When you’re doing kind of deeper reading stuff, they may get there. I mean, the AIs may know everything at some point, but like there’s not a super strong reason for them to get there earlier. So there are pockets like that that probably will be specialized for longer. But, you know, as a school, it’s just better for us the faster all of that becomes a commodity.

And the more we can just, you know, get off-the-shelf stuff, like there’s no real joy in building all of this stuff. And for the change to happen, we don’t want people to have to think about all this stuff, really.

Diane Tavenner: No, I have to ask about scale because your point that the faster we can get kids to be living in today’s world versus the old world suggests that we need to scale as quickly as possible for that to happen, to get as many kids there. You and I both bear a lot of scars around different efforts to scale both mortar schools and influence type things. This time you’ve gone with a microschool network. What’s your, you had grand ambitions with Rocketship and clearly Rocketship’s great and Preston’s done an amazing job since you left, but it never reached sort of the scale that I think you originally hoped. What is your thinking now? Why microschools?

John Danner: Yeah, I mean, you know, putting it like just putting it bluntly, I think politics killed charter schools more or less. Like, you know, you look at most high-performing charter schools, they tend to look more and more like the districts that host them. You know, they actually, like, I look at RocketShips around the country. They actually look as much like the district they’re hosted by as they look like RocketShips sometimes. You know, it’s like, ’cause you know, your authorizer authorizes you and they have a lot of influence. So it was kind of like this cool experiment that at the beginning probably created a lot of innovation and then over time kind of has this like bringing it back to the, you know, kind of what the districts are doing. I think that microschools, certainly microschools, are starting in a very different place, you know, where the way I think about charters is the compromise happened right at the beginning. Where we would like to receive public funding and for that we will like to fit into the system.

Whereas the microschool movement kind of started with a different point where the stronger position was taken early on when the laws were formed that like these things are independent. They’re way more like private schools than they are like district schools. And of course, there will be some influence from states and others on that, but nowhere near like, you know, what we saw in the charter world where it was like, you know, I remember the story I always tell is Rocketship had specialized teachers for math and reading in elementary school, which was not normal at all. And I was just tortured for years by districts over this. You know, the main thing was like, no, it’s, you know, a student needs one trusted adult, you know, when they’re that age. And if they have two, it’s going to like, you know, all fall apart, which was, of course, total bogusness. But I had to go through that anyway. Like, you know, that was just time of my life spent arguing something silly.

Whereas with microschools, you just don’t have to argue that. So I think the big question is, what will be the ultimate, like, kind of political destiny of microschools? Will they get capped in the way that charters did? Will they somehow kind of get influenced in a way they aren’t now? Right now they’re pretty great. I mean, you know, you basically build a school that parents and students love and, and you build the curriculum and the program you want. That’s nice. Something you would have enjoyed, Diane.

Reimagining Teachers’ Roles

Diane Tavenner: Yeah, no, I mean, it’s tempting. I will say Michael’s always so kind because when we start talking schools, I just take over. So he’s being so patient. The thing that’s coming to me, and maybe this will lead us to wrap up, is, you know, you and I both taught, and were passionate about teaching. And as you start talking about politics, one of the sort of sad elements of that politics to me is I think teachers get involved in kind of, or, you know, blocking some of these changes, a lot out of fear, a lot of out of like but my identity is teaching a classroom of students and writing great curriculum and like doing all, you know, being a hero. And I think what you’re offering is a new identity for a teacher that might actually be more aligned with why they got into it in the beginning, which is instead of judging myself by the quality of my classroom instruction, I’m like literally focused on every single kid learning and growing and, you know, in your words, flourishing, right? It’s such a profound

John Danner: In general, I think that professions that go in the direction of being more human, where the human elements are like the differentiator, they’re going to do so much better. So I, you know, wrote a piece on this. I just think, you know, while most parents would not have counseled their kids to become teachers in the last 20 years, I think that conversation is likely to change because I think it’s going to be both a more enjoyable job and probably more resilient to kind of the whole AI apocalypse than most jobs.

Michael Horn: Agreed.

John Danner: Yeah.

Michael Horn: I think that is a good place to part us. But John, I feel like we have like 10 other questions like sitting in our dock that we could have dug in with you. But let’s pivot. This is fascinating. It’s really cool to see what you’re building and hear both the frustrations, but also frankly, the North Star for where it’s going. And one day maybe Massachusetts will have you here. But I’ll pray for now. But let’s pivot.

This season of Class Disrupted is sponsored by Learner Studio, a nonprofit motivated by one question. What will young people need to be inspired and prepared to flourish in the age of AI as individuals, in careers and for civil thriving? Learner Studio is sponsoring this season on AI in Education. Because in this critical moment, we need more than just hype. We need authentic conversations asking the right questions from a place of real curiosity and learning. You can learn more about Learners Studio鈥檚 mission and the innovators who inspire them at www.learnerstudio.org.

We have this section that we always talk about things we’re reading, watching, listening to. We try to do outside of work. People track us on this stuff. Diane and I occasionally fail. I’m going to fail today. So you can go wherever you want.

John Danner: So, yeah. I’m rereading the Culture series, Iain Banks, right now. So my brother works for Tesla and Tesla just, as you probably heard, kind of made this transition where they knocked off the Model S and Model X and are building robots. So he’s building robots right now. So that makes it much more personal to me that like the future is coming soon, and so, you know, I’ve always been a science fiction reader, but, but I think one of the cheat codes in Silicon Valley is like the amount of science fiction consumed equals your ability to be comfortable with like what’s coming. So yeah, culture series.

Michael Horn: Good rec, good rec.

Diane, what’s on your list? You said you’re cheating.

Diane Tavenner: So, I’m cheating, I’m failing today. Sorry. Ted Dintersmith has his latest book out and sent it along. I couldn’t resist. The title is very provocative. It’s called Aftermath: The Life-Changing Math That Schools Won’t Teach You. And, you know, this is really, you know, for those who don’t remember, Ted, like, goes hard on the things we’re doing wrong and really tries to bring public awareness to them. And, I think lots of us have been concerned about how math is taught and not taught and whatnot for a long time.

So, that’s what this one’s about.

Michael Horn: I have an email from him in my inbox to send him my address, so I will do it after this conversation, uh, so he could send it to me as well. But, I’m also cheating. I’ve been really interested in, not just how schools start doing new things, but how do they stop doing old things? Like, they are just really bad. And it’s not just schools, by the way. Like, all organizations are really bad at deimplementing or pruning, like, old things that don’t make sense anymore, whether they’re bad habits or frankly habits that just aren’t fit for the current age. So I’ve started, like, trying to read some of the academic literature and just learn about that. And there’s a book, Making Room for Impact: A Deimplementation Guide for Educators, by Aaron Hamilton, John Hattie, and Dylan William. And so I’m just cresting the end of that book right now, and, and then looking at all the healthcare studies that they’re citing.

And I haven’t decided if I’m going to read those, but that’s where I am right now.

Diane Tavenner: So is it a recommend, Michael, or no?

Michael Horn: I mean, it’s, it’s like a, it’s a deep workbook, right, on the topic, um, is what I would say. So like, if you’re a school and you’re trying to work through this, definitely dive into it. I was more interested in like, who’s, who’s thought about, like, how do you de-implement? How do you prune, right? And because there’s just not a lot of conversation except for educators griping about it. And so I wanted to learn more and it was a good starting point. So huge thanks, John, again for joining us. We appreciate it. Really check out his Substack as well if you want to just sort of follow along on the journey, I guess is what I would say. And we’ll watch as Flourish opens two more in Arizona in August and keep up the good work.

We appreciate you. And for all of our listeners, keep the emails, notes coming. We love it. We learn a lot from it as well, and it inspires us on our future topics. And so, as always, thanks for joining us on Class Disrupted. We’ll see you next time.

This episode is sponsored by LearnerStudio.

In this one-hour webinar, hosted by AERA and 社区黑料, leading education researchers will explore how AI is being used in assessment today, what evidence we have about its effectiveness and what risks demand careful attention. The conversation will balance promise with caution, highlighting both cutting-edge research and the policy and ethical considerations shaping the future of student assessment.

RSVP to watch, or refresh after the webinar to stream.

Related coverage on 社区黑料: 

• Two New Reports Urge 鈥楬uman-Centered鈥� School AI Adoption
  
• AI Optimization鈥檚 Impact on Use of Time, Space and Resources in Schools
  
• 10 Useful Tech Tools for Educators in 2026: A Practical Guide 鈥� 社区黑料

Updated March 27, 2026:听In response to this story, YouTube terminated six channels for violating the platform鈥檚 terms of service and one channel for violating its spam policy.

In a video that has been played almost 50,000 times since it was posted five months ago, two cartoon children sing along as they guide viewers through the experience of riding in a car amid a vividly colored, utopian backdrop. 

At first, the seems harmless. The song is upbeat and informative. The animation aligns with the promised subject. 

Except, hold on a second, did those lyrics just say, 鈥淩ed means stop, and green means right鈥�? And why are the characters changing in every frame 鈥� different hairstyles and colors, slightly different outfits for the girl and boy? 

Worst of all, for a video that purports to be 鈥渆ducational,鈥� the visuals are sending precisely the wrong message about riding in a car. 

The video opens with the children riding, without seatbelts, in the front row of a moving vehicle. The next scene shows the girl defying physics, floating alongside a moving car, while the boy is seated in what appears to be the hood of the vehicle as it travels backward down a busy street. The third and fourth scenes show the children walking in the middle of the road with moving cars behind them. 

It鈥檚 not hard to imagine how the video could have gotten so many views. 

Maybe a parent needs to complete a task 鈥� fold some laundry, get dinner ready, hop in the shower 鈥� and is searching for an age-appropriate video on YouTube to entertain their toddler during that short time. Perhaps that toddler, increasingly independent and prone to running off, needs a better grasp of road safety. 鈥淰room Vroom! Car Ride Song | Educational Nursery Rhyme for Kids鈥� presents itself as a win-win solution. 

But children鈥檚 media experts say this is AI-generated 鈥渟lop,鈥� and that it has infiltrated the internet, preying on young children and their unsuspecting caregivers. 

鈥淲e鈥檙e at the beginning of a monster problem, and we have to get hold of it quickly,鈥� said Kathy Hirsh-Pasek, a professor of psychology and neuroscience at Temple University and senior fellow at Brookings Institution who studies child development. 

She and other researchers, including Dr. Dana Suskind, a professor of surgery and pediatrics at the University of Chicago, have that AI-derived products for babies and children need to be reined in. 

鈥淭his is not neutral content,鈥� said Suskind, author of the forthcoming book . 鈥淚 think of this as toddler AI misinformation at an industrial scale. It鈥檚 very risky for the developing brain.鈥�

It鈥檚 hard to say just how pervasive this type of content is, but it鈥檚 clear the problem is widespread and getting worse. One published by video-editing company Kapwing in November 2025 found that about 21% of YouTube鈥檚 feed consists of low-quality, AI-generated videos. 

, the creator of the 鈥淰room Vroom! Car Ride Song,鈥� has posted more than 10,000 videos since its first release just seven months ago, in August 2025. That鈥檚 an average of about 50 new videos each day. , meanwhile, has published about 3,900 videos to YouTube in its entire 20 years on the platform. 

The cognitive decline associated with the consumption of AI slop 鈥� such as a shortened attention span, decreased focus and mental fog 鈥� is sometimes referred to as 鈥渂rainrot.鈥� But when the audience is children, there鈥檚 not much to rot, Suskind said. Because a child鈥檚 brain is still in its early development, still being built, what you get instead, she said, is 鈥渂rain stunt.鈥�

鈥淓very experience is building a million new neural connections,鈥� Suskind said of children who are still in their early years. 鈥淵ou will be unintentionally wiring the brain in incorrect ways.鈥�

That comes at a cost. A child may absorb the implicit messages of something like the Vroom Vroom video and end up mimicking the 鈥渄ownright dangerous鈥� behaviors they saw depicted there, said Carla Engelbrecht, who has created digital experiences for children鈥檚 media brands such as Sesame Street, PBS Kids and Highlights for Children and considers herself an AI educator and creator.

Engelbrecht is also when it comes to child-targeted AI slop. She has found countless examples of AI-generated videos that could cause real physical harm.

鈥淭he more content I find,鈥� she said, 鈥渢he more horrified I get.鈥�

They include videos of a being chased by a T-Rex; a crawling biting into an apple that appears bloody, swallowing whole grapes (a major) and eating honey (which carries the potentially fatal risk of ); and a eating raw elderberries (which are toxic when uncooked).

But there鈥檚 another category of AI slop in kids鈥� media, she said, with consequences that are more difficult to capture. These videos claim to pertain to learning and development, focusing on topics like literacy and numeracy, but due to the speed with which they are produced and the lack of quality checks, they end up introducing or enforcing the wrong lessons. And sometimes, the errors don鈥檛 come until midway through the content. That means if a parent previews the first few seconds of a video, they may miss the unreliable information that appears later in the clip.

A about vowels includes visuals of consonants. It also depicts letters on screen that don鈥檛 align with the audio overlay. A promising to teach about the 50 U.S. states sings along as butchered state names appear in text at the bottom of the screen 鈥� Ribio Island, Conmecticut, Oklolodia, Louggisslia. A about the seven continents frequently shows a compass with more than four points and indecipherable symbols where the 鈥淣,鈥� 鈥淪,鈥� 鈥淓鈥� and 鈥淲鈥� should be.

These may seem like silly slips from a machine, but for a child, every 鈥渋nput鈥� is part of their learning process, Engelbrecht explained. 鈥淢ixed signals means you are delaying them learning the cause and effect of a thing,鈥� she said. 鈥淚f you learn that red is blue and blue is red, that鈥檚 a delay.鈥�

鈥淚f you鈥檙e inconsistent, it takes that much longer to learn,鈥� she added. 鈥淓very delay they have means everything else gets pushed back. That鈥檚 taking their executive function offline to go learn nonsense.鈥�

Amid all of this internet muck, the question of responsibility is a tricky one.

鈥淔undamentally, everybody has a responsibility,鈥� Engelbrecht said, including platforms like YouTube; companies that operate large-language models, like OpenAI, Google and Anthropic; the people creating and publishing these poor-quality videos intended to reach kids; and parents. 

YouTube鈥檚 current requires creators to disclose videos that have been generated by or altered with AI when that content 鈥渟eems realistic.鈥� This does not apply to cartoons and 鈥� which seems to be the majority of what鈥檚 reaching children 鈥� because it has long been assumed to be fictional content, Engelbrecht explained. 

The platform does have stricter 鈥溾�� for content targeting children than it does for its general viewership, said Boot Bullwinkle, a YouTube spokesperson, in a statement. It also has a 鈥�.鈥� (These web pages, however, do not specifically address the use of AI.)

Due to the volume of content on the platform, YouTube does not catch every video that violates its policies. (It did take action against at least seven channels on the platform in response to 社区黑料鈥檚 reporting, including terminating two.) 

鈥淭he trust that parents and families put in YouTube is a responsibility we take very seriously, and we鈥檝e invested deeply in age-appropriate environments that empower parents,鈥� Bullwinkle wrote in the statement. 鈥淵ouTube Kids, for instance, offers industry-leading parental controls and rigorous designed to provide a safer experience for families.鈥�

YouTube Kids is a distinct version of the platform with content that has been curated for children from birth to 12. Many families continue to use the main YouTube platform to view children鈥檚 content, though, which means many creators still have an audience and earning opportunities there. None of the AI-generated videos reviewed for this story were found on YouTube Kids, although recent in The New York Times found AI videos had penetrated that space as well.

Sierra Boone, executive producer of Boone Productions, a children鈥檚 media production company that makes original content for children ages 2 to 6, noted that kid-friendly competitors to YouTube, such as by Common Sense Media and , do exist. But they have struggled to break through to families. 

鈥淥vercoming that juggernaut is extremely difficult,鈥� Engelbrecht said of YouTube. 鈥淭here鈥檚 a graveyard full of failed attempts to create a safe YouTube alternative.鈥�

Boone suggested that some effective labeling would go a long way, not unlike the 鈥溾�� LinkedIn is phasing in, which aim to disclose when media has been created or edited by AI, in part or in whole. 

Engelbrecht thinks labels are a good idea, not least because they would be important for AI literacy, but she also believes they would penalize creators like her who use AI 鈥渢houghtfully鈥� in their work. (She is , among other projects, an AI tool that detects AI slop in children鈥檚 videos on YouTube.)

As for who鈥檚 behind the videos, some of it originates overseas, but plenty is home-grown, created by Americans with access to phones or computers who are just trying to 鈥渕ake a quick buck,鈥� as Boone put it. 

These people are often using AI at every step of the process 鈥� to develop themes and scripts for children鈥檚 videos, to generate the videos, and to automate the process of publishing the content regularly on 鈥�, in which the creator is anonymous and has no on-camera presence, Engelbrecht explained.

A little over a year ago, a popular content creator posted a video to YouTube in which she raves about a 鈥渉uge opportunity鈥� that would lead to 鈥渕any millionaires.鈥� The opportunity? AI-generated animated videos that inexperienced users could create with a simple prompt in just minutes. The target audience? Young children. 

That video has been viewed more than 335,000 times. 

鈥淎I in general isn鈥檛 inherently good or bad, but it exposes people鈥檚 intentions,鈥� said Boone, whose production studio is responsible for . 

The flood of AI-generated content, she added, reveals how many people have 鈥渘o regard for children or how they鈥檙e impacted,鈥� as long as it benefits them. 

For Boone, who works painstakingly with her team on every episode of The Naptime Show 鈥� researching, writing the script, editing the script, placing props, doing table reads, going to set, filming, editing the video, publishing and promoting the final product 鈥� creating children鈥檚 media is an 鈥渉onor鈥� that should be taken seriously. 

鈥淭he very foundation of creating children鈥檚 media is you are creating something that a child, in their core developmental years, is going to be consuming,鈥� Boone said. 鈥淪o what is the level of intention that you鈥檙e bringing to that? I think we need to be holding the people who are uploading this content more accountable.鈥�

Ultimately, though, in the absence of more regulation or content moderation, the burden falls on parents. 

Parents are likely putting YouTube videos in front of their children in the first place because 鈥渢hey are already so stretched,鈥� said Suskind, who still sees patients in her pediatric practice and interacts with families often. So it鈥檚 inherently challenging to ask them to more closely monitor the content that is coming through their children鈥檚 screens. 

Yet that is what must be done, Hirsh-Pasek said. Until a better solution emerges, the onus is on parents to separate the slop from 鈥渢he good stuff.鈥�

鈥淲e owe it to our kids to protect them,鈥� said Hirsh-Pasek. 鈥淭hat鈥檚 what they look to parents for, to keep them in safe spaces. If we don鈥檛 deal with that or do anything about that, we鈥檝e absconded [from] our responsibility.鈥�

New York City students with a passion for STEM 鈥� and an interest in artificial intelligence 鈥� may soon have a high school dedicated to training 鈥渢he next generation of technology professionals.鈥�

But families in Manhattan鈥檚 District 2 are pushing back against for , a new screened admissions high school that would take the place of the tiny, girls-only Urban Assembly School of Business for Young Women. Next Generation would be the first city public school to focus its curriculum on AI and computer science.

As details of the two proposals emerged over the last month, so have dual tensions: What should fill the space left by Young Women in Business, and how private technology companies and their artificial intelligence products could shape the curriculum at Next Generation.

Much of the opposition to Next Generation has come from families at a middle school also in the Broadway building, Lower Manhattan Community School. Also known as LMC, parents at the school have called on the department for years to expand enrollment from grades 6-8 up to grade 12.

The Panel for Educational Policy, the board that votes on new schools and closures, is expected to consider the proposals for Next Generation and Business for Young Women at its April 29 meeting.

The Education Department released both proposals on March 6, the day after the city鈥檚 eighth graders received their high school acceptance offers. If approved, Next Generation would welcome its first class of ninth graders in the fall. (The plan to close Business for Young Women in June is not contingent on Next Generation鈥檚 approval.)

Despite not having the green light yet, Next Generation has already held three virtual open houses. Its states the school is 鈥渟et to open鈥� in fall 2026, noting that applications would open March 19.

Parents ask: 鈥榃hy this school and why here?鈥�

Manhattan High Schools Superintendent Gary Beidleman introduced the idea for Next Generation Technology High School at a .

Panel for Educational Policy members and families of the three co-located schools at 26 Broadway 鈥� in addition to LMC and Business for Young Women, Richard R. Green High School of Teaching shares the building 鈥� said that meeting was the first time the district school community had been notified of the proposed STEM- and technology-focused screened high school.

At the Feb. 25 announcement, Beidleman said Next Generation grew out of his experience as a summer 2024 , and that Google and OpenAI are part of the planning team for the school. One of the school鈥檚 goals, he said, is to 鈥渆xpand pathways connected to high-growth technology careers鈥� and provide advanced STEM and technology programming for NYC students. Next Generation also plans to offer a summer internship program with Carnegie Mellon University.

Caleb Haraguchi-Combs, founding principal and project director of Next Generation High School, said in an information session that the school would utilize . How much of this AI-powered, AI-focused Google coursework would comprise the curriculum is still in flux, according to the proposal鈥檚 .

The school鈥檚 academic description includes similar or identical language as found on the Google Skills website: Next Generation鈥檚 鈥渟pecial access to technology industry mentors,鈥� 鈥渢echnology certifications,鈥� and 鈥渃urriculum that adapts to the dynamic changes in the technology field鈥� are offerings advertised on the homepage of the Google Skills site.

Officials and families question new school proposal process

The community and Panel for Educational Policy members have asked questions about the fast proposal process, speaking to uncertainty around admissions for the coming school year.

in a letter to the Panel for Educational Policy that the proposal seemingly came out of nowhere, and families were not provided adequate engagement opportunities before its release. Panel Chair Greg Faulkner said he has received hundreds of similar letters from parents since the community learned of the incoming proposal in late February.

High school offers were released March 5, ahead of the panel鈥檚 vote and months before the proposed school would open. It remains unclear how the Education Department would handle screening requirements 鈥� such as interviews or assessments 鈥� after the main admissions cycle has concluded. The Office of District Planning did not respond to questions about how enrollment would work for this fall.

of the school, created by the Next Generation鈥檚 founding principal and program director on March 8, had under 100 signatures at the time of publishing.

A public hearing is scheduled for April 14, two weeks before the panel鈥檚 vote.

鈥淚 would love more transparency around why the department chooses certain schools to go in certain places,鈥� said Sarah Calderon, a parent at Lower Manhattan Community School. 鈥淲hen we asked the superintendent, 鈥榃hy this school and why here?鈥� he said he had no data on district demand.鈥�

Beidelman told parents at the Feb. 25 District 2 meeting that expanding Lower Manhattan Community 鈥渨as not an idea that was on the table.鈥�

The Education Department receives many proposals each year, including some from outside New York City, said Sean Rux of the Office of New School Development.

鈥淭his was the proposal that spoke to us,鈥� Rux said.

Families push to expand Lower Manhattan Community School

The plan to close the underenrolled Business for Young Women school has been percolating for a few years 鈥� with just 91 students this year, it鈥檚 the smallest district high school in the city, said Education Department officials.

Families at Lower Manhattan Community School say they have pushed for years to expand into a 6鈥�12 model, and would like to move into the space used by Business for Young Women, if closed.

鈥淎 proposal to expand LMC could potentially open up sixth grade admissions to applicants citywide, but we have not been given the opportunity to even submit a proposal,鈥� said Anne Hager, a parent of a sixth grader at Lower Manhattan School.

At a PTA meeting with Education Department staff on Wednesday, LMC鈥檚 Student Leadership Team presented its case to expand the school instead of opening Next Generation.

A new 6-12 would eliminate the need for LMC students to go through a second, onerous application process, something that students with disabilities would especially benefit from, they said. The presentation also cited Department of Education data from 2024 that showed 6-12 schools have nearly three times higher demand than their 6-8 middle school counterparts.

compared with citywide averages.

The department鈥檚 proposal focuses largely on space at the Broadway campus, estimating that Next Generation would serve roughly 450 students by its fourth year. All three schools can comfortably co-locate, according to the proposal, though its capacity calculations do not allot for significant expansion for either Richard R. Green High School or LMC.

Debate over AI timing and oversight

Next Generation鈥檚 proposal arrives amid over artificial intelligence in schools.

The school initially marketed itself in information sessions and on social media as an 鈥淎I school,鈥� though DOE officials later clarified that students would learn about artificial intelligence rather than be taught by it.

鈥淪tudents need to be creators, not consumers, of technology,鈥� Beidleman said at the Feb. 25 meeting. 鈥淟essons learned from the past show us that new tech in place creates an opportunity.鈥�

Some parents have argued that broad use of an AI platform in public schools should not be allowed before comprehensive guidelines have been released by the city.

Greg Faulkner, who chairs the Panel for Educational Policy, said he first learned of the proposal after receiving Next Generation鈥檚 last month. Since then, the panel has received hundreds of letters from parents opposing the plan and raising concerns about the lack of community engagement so far.

鈥淚 have two major hesitations with this: We don鈥檛 know what kind of AI involvement there will be. The development team has not provided a playbook for how that will look,鈥� Faulkner said. 鈥淎nd in reading the response letters from District 2 parents, I see that proper engagement and process was not done.鈥�

At a District 2 town hall on March 5, Chancellor Kamar Samuels said the Education Department expects to release AI guidance in the coming weeks and will provide a 45-day window for community feedback once it鈥檚 published.

Five Community Education Councils have passed resolutions calling for a two-year moratorium on artificial intelligence use in schools. But calls for broad AI guidelines implemented at the city level are nothing new; of an AI-powered reading program in 2024 after former Comptroller Brad Lander called for a citywide playbook.

鈥淚 think the question of teacher capacity and teacher shortages, the research on kids and AI, is still nascent, and the DOE鈥檚 lack of its own AI policy leads me to question the timing of any AI school,鈥� said Calderon, the parent at Lower Manhattan Community.

Chalkbeat is a nonprofit news site covering educational change in public schools. This story was originally published by Chalkbeat. Sign up for their newsletters at .听

Months before the Los Angeles school board approved a $6.2 million contract with AllHere, an AI chatbot maker that is now being investigated by the FBI, top district leaders were invited to a meeting with its CEO and a consultant, who is a close friend and associate of schools Superintendent Alberto Carvalho.

The Jan. 18, 2023, calendar invite for the gathering at the district鈥檚 downtown headquarters, billed as 鈥淎llHere Meeting,鈥� was shared with 社区黑料 by a former central office staffer, who asked to remain anonymous for fear of retribution. 

The AllHere contract in question is widely believed to be connected to the high-profile raids on Carvalho鈥檚 home and district office in late February. 

社区黑料 has not received confirmation on whether the meeting took place or what specifically may have been discussed, but the invite suggests district administrators were consulting with AllHere principals five months before the contract was voted on.

It also calls into question public statements by Carvalho, who was placed on paid leave Feb. 27, that he . He said the education technology venture represented by his longtime friend and business associate Debra Kerr won the job based on legally mandated bidding. Kerr called the Jan. 18 meeting.

AllHere filed for bankruptcy in September 2024 and its founder and CEO, Joanna Smith-Griffin, was later arrested on charges of identity theft and defrauding investors. 

社区黑料 filed extensive public record requests with Los Angeles Unified School District in September 2024 for documents related to the AI chatbot contract, including all proposals, bids or submissions made by AllHere and any other companies vying for the work. The request also asked for documents detailing how the district evaluated AllHere鈥檚 qualifications and determined that the small Boston-based firm with little to no artificial intelligence experience was capable of carrying out the contract.

On Feb. 11, 17 months after those requests were filed and two weeks before the FBI raids, a senior paralegal in sent 社区黑料 an email asking if we still wanted the documents.

Through his attorneys and a spokesperson, Carvalho since the FBI probe exploded into public view. The Los Angeles Times reported that he denied any wrongdoing, pointed out that “no evidence has been presented by prosecutors supporting any allegation that (he) violated federal law” and pressed to return to his job.

鈥淢r. Carvalho remains confident that the evidence will ultimately demonstrate that he acted appropriately and in the best interests of students,鈥� said the statement that was issued through the spokesperson and the law firm of Holland & Knight, according to the Times. 鈥淲e hope the school board reinstates him promptly to his position as superintendent.鈥�

Kate Brody, the vice president of communications for , a 2,000-member LAUSD parent and educator advocacy group, sees the moment differently. Her group has called for an audit of all the education technology contracts entered into under Carvalho, saying they lack independent research into their efficacy and now is “the time to peel this whole thing back and take a look, not just at what鈥檚 going on with AllHere, but the inappropriate amount of access that all these companies have.”

“The evidence is increasingly clear that this technology is not really for the benefit of the students,” she told 社区黑料. “Our big question has been for a long time 鈥� whose benefit is it for?”

Carvalho has not been accused of any wrongdoing and authorities have not provided details about the investigation. The warrants underlying the . 

In  after the Board of Education placed Carvalho on paid leave and named an acting superintendent, the district said that while it understood 鈥渢he need for information, we cannot discuss the specifics of this matter pending investigation.鈥�

Kerr could not be reached for comment and attorneys for  Smith-Griffin did not respond to requests for comment. District spokesperson, Britt Vaughan, could not be reached for comment.

Kerr and Carvalho

Federal agents also . Her ties to Carvalho go back to his days leading the Miami-Dade County Public Schools, a period of time in his prominent career that is also now reportedly under investigation. According to , grand jury subpoenas have been issued seeking records from the district鈥檚 inspector general and a fundraising foundation overseen by Carvalho while he was the Miami schools chief.

Kerr was a key player in executing the failed contract between AllHere and the nation鈥檚 second-largest school district. In addition to her being in a position to call senior staff to a meeting at district headquarters, according to the calendar invite, Kerr鈥檚 son Richard, a former AllHere account manager who began working for the company in 2022, told 社区黑料 in September 2024 he pitched AllHere to LAUSD school leaders.

Among 社区黑料鈥檚 long-unanswered public records requests were any conflict of interest disclosure forms filed by AllHere, its employees, third parties involved in the contract or LAUSD personnel.

The location listed on Kerr鈥檚 hourlong invite to discuss AllHere was the office of LAUSD鈥檚 longtime chief spokesperson Shannon Haber, who has since retired. Other invitees included senior advisor of communication B铆ch Ng峄峜 Cao, senior director of engagement and partnerships Antonio Plascencia Jr.. and director of development and civic engagement Sara Mooney. 

Mooney is also the former executive director of the , the district鈥檚 separate fundraising arm includes Carvalho. Attempts to reach Haber and the other meeting invitees, which also included Vaughan, the district spokesperson, and marketing director Lourdes Valentine, were unsuccessful.

Earlier calendar entries shared with 社区黑料 show Carvalho had an hourlong meeting scheduled with Kerr and someone identified only as 鈥淪N鈥� on Oct. 21, 2022, about eight months after he took the $440,000-a-year job in Los Angeles. The meeting was scheduled for 12:30 p.m. at a place 鈥渢o be determined.鈥�

In 2022, Kerr was busy consulting for and promoting AllHere in multiple Florida cities, according to . She also did consulting work for Rethink Ed, a New York-based company that provides social-emotional and wellness resources. In May 2020, in the midst of the COVID-19 pandemic and the national school shutdowns, to support students with autism and other related disabilities during remote learning. 

“We appreciate partners like Rethink Ed which assist us in empowering these very deserving students with a variety of innovative and helpful tools to successfully engage in distance learning,鈥� Carvalho said in a statement when the Miami-Dade contract was announced.

Roughly two years later, when Carvalho was leading LAUSD, the firm

Other calendar entries shared with 社区黑料 show that right before the scheduled meeting with Kerr that October Friday, Carvalho had back-to-back interviews lined up with reporters from The Wall Street Journal and Politico. Later that day, he was scheduled to attend a retirement dinner for Michael Hinojosa, the former Dallas schools superintendent, at the Ravello restaurant at the Four Seasons in Buena Vista Lake, Florida, near Orlando.

Two days before Carvalho was due back in Florida for that celebration, the a $1.89 million contract to provide text-messaging support to students struggling with attendance, academics and social-emotional issues. The SMS tool was a precursor to its AI-powered chatbot. 

Carvalho told the Los Angeles Times he had getting the three-year deal in Miami although the newspaper reported that the bidding process began while he was still in charge. 

Two years later, in November 2024, the district would move with Miami-Dade schools for a period of three years after the ed tech company abandoned its contract.

社区黑料 filed public records requests on Sept. 13, 2024, asking for copies for all of Carvalho鈥檚 daily calendars going back to his first date of employment at LAUSD. The district has yet to produce them.  

AllHere then gone

Also invited to the Jan. 18, 2023, meeting set up by Kerr was AllHere鈥檚 Smith-Griffin, who six months after landing the L.A. schools deal was charged with defrauding investors of nearly $10 million.

Her case, which involves allegations of securities and wire fraud and aggravated identity theft, is being heard in U.S. District Court in Manhattan. The Harvard graduate and former middle school math teacher  pleaded not guilty in December 2024. Conferences on her case were postponed three separate times in 2025 to allow the parties time to work on a possible disposition. The last was a 60-day adjournment on Sept. 25, 2025, and there鈥檚 been no activity in the file since then.

By the time Smith-Griffin was arrested at her home in Raleigh, North Carolina, in November 2024, the company she founded in 2016 had been forced into bankruptcy, unable to pay its debts, including a disputed $630,000 commission claimed by its largest creditor: Kerr.

Carvalho and Smith-Griffin spent considerable time together in the spring of 2024, appearing at multiple ed tech conferences touting 鈥淓d,鈥� their sunny chatbot that was seen as catapulting LAUSD into the K-12 AI vanguard. They said communicating with Ed would provide an unprecedented level of support, accelerating learning and strengthening well-being for students and families, many of whom were still struggling from the pandemic. 

鈥淗e鈥檚 going to talk to you in 100 different languages, he鈥檚 going to connect with you, he鈥檚 going to fall in love with you,鈥� Carvalho raved at the April 2024 ASU+GSV conference in San Diego. 鈥淗opefully you鈥檒l love it, and in the process we are transforming a school system of 540,000 students into 540,000 鈥榮chools of one鈥� through absolute personalization and individualization.鈥�

None of that materialized for the district, whose enrollment has since and which is now and

After AllHere shuttered and a former company manager-turned-whistleblower told 社区黑料 that students鈥� private data  was not properly protected in the push to launch Ed, Carvalho vowed to investigate. He promised a task force of outside experts who would dig into what went wrong with the AllHere contract and determine how the district could strengthen its bidding process to avoid future debacles.

Carvalho told the Los Angeles Times in July 2024, he expected. Some 19 months later, there鈥檚 been no further news or shared task force findings. The district鈥檚 independent inspector general鈥檚 office launched its own investigation around the same time. 

However, the office鈥檚 and reports to the Board of Education make no mention of AllHere. In 2024, the IG opened a total of 62 cases, closed 54 and identified nearly $2.5 million in waste. In 2025, it opened 38 cases and closed 43, including some from previous years, though none appear to have involved AllHere. No financial waste was identified in 2025. 

Inspector General Sue Stengel at the end of 2025 after three years. The office did not respond to a request for comment. 

Equally elusive is what happened to Ed or the underlying tech tool for which LAUSD paid AllHere $3 million out of its $6.2 million contract. Although it鈥檚 been reported that school officials said the district was not financially harmed in the contractual fallout, and it received the services and products it spent several million dollars to acquire, it鈥檚 difficult to substantiate that.

When Carvalho unveiled Ed at a major March 20, 2024, celebration attended by Gov. Gavin Newsom and L.A. Mayor Karen Bass, he said the chatbot would be in 101 elementary, middle and high schools as part of a pilot program. By the fall, Ed was supposed to go districtwide. 

Much later, that reported group of Ed testers had been 鈥渢o a small number of schools (that) tried it out, each with a sample of students and parents.鈥� In July 2024 after the district 鈥渦nplugged鈥� Ed in the wake of AllHere鈥檚 demise, that it was 鈥渉ard to find students, teachers or other staff who have used any part of the system since its official launch.鈥� 

Absent human interactions with Ed, the district has been slow to produce documentation from AllHere of services rendered. Among the public records sought by 社区黑料 in September 2024, which LAUSD now appears ready to provide, are 鈥減urchase orders, invoices, and payments records related to any and all goods and/or services provided by AllHere.鈥� 

Staff reporter Amanda Geduld contributed to this report

The private, AI-powered Alpha School had quickly generated attention in the education world and beyond. The school鈥檚 been featured in dozens of articles and dissected across countless podcasts for what leaders call their 鈥渢wo-hour learning鈥� model.

On this episode of Class Disrupted, MacKenzie Price, co-founder of the Alpha School, joins Michael Horn and Diane Tavenner not to explain Alpha School鈥檚 model, but instead to dive deep into how the school is leveraging artificial intelligence to radically rethink the school experience. Price focuses on how AI itself is being leveraged at Alpha 鈥� from the core academic blocks to afternoons spent on real-world projects and life skills development. What鈥檚 possible now in school design that wasn鈥檛 a decade earlier, thanks to AI? 

Listen to the episode below. A full transcript follows.

Chris Hein: So when the school shut down and went to remote learning, we were really fascinated by how quickly our kids adjusted to e-learning and how hard of a time the teachers seem to have with just the basic tools and systems and then how to translate their curriculum to a digital format. But the thing that really jumped at me was my wife and I were having conversations with our kids every day saying, hey, what are you doing?

Why are you guys playing video games? Or why do you, like, want to go outside and play? It’s midway through the day and they’re like, we’ve already done our work. And we were like, that can’t be right. And so we double checked their assignments and their tests and where they’re at. And it was like, no, they got all their work done in a couple hours. And then it really made Teresa and I question, why does it take them eight hours a day at school if the school is teaching them the same content and administering the same number of tests and they’re able to get through it in a few hours?

Michael Horn: That was June 2020, and Diane and I were broadcasting during the height of the pandemic, and we were hoping that parents would realize that schools could be rethought dramatically, including by helping people realize that what we tend to think of as, quote, the academics could be done in much, much less time than the six plus hours that kids spend in traditional schools. Five years later, and thanks to a startup school network, Alpha School, the two hour message finally seems to be spreading like wildfire. So with that as a prelude, Diane, first, it is great to see you as always.

Diane Tavenner: It’s good to see you too, Michael. I’m a little disoriented by us changing up our normal intro. But in a good way, change is always good. That take from season one is honestly priceless. It’s taken us a bit longer than we had hoped, but we do seem to be getting some momentum towards some of the big opportunities that we saw in education back then and still are hopeful for now.

Michael Horn: Yeah, no, I think that’s right. And I’m glad you’re accommodating my whims on changing the format up on you today. But I am particularly excited because we have on our show today MacKenzie Price. She’s of course one of the co-founders of Alpha School, and MacKenzie’s been on my Future of Education podcast and Substack before and we actually both have Substacks named the Future of Education. We independently named them, so we’re vibing already. But MacKenzie, it’s great to see you again, welcome.

Scaling Education with Technology

MacKenzie Price: Well, thanks for having me. And, you know, it’s so interesting that you tell that story about the way, you know, education was done during COVID And we were pretty lucky because we’d started Alpha school back in 2014. So when the pandemic hit, you know, it happened to be during spring break. So the kids who hadn’t brought their laptops home came and picked them up at school. And we really had a very smooth rest of the school year because the kids already were doing their learning on the computers. And then we just said, you know, afternoons, we’ll just, we’ll, we’ll call it, you know, do whatever you want at home. But what’s interesting is a couple years ago or in 2022, when we really launched our learning platform with the advent of generative AI, we realized, okay, we can actually scale this. We can go beyond just, you know, a local school that’s doing a reasonable job of educating kids, and we can, we can scale it bigger.

And we were originally talking about the idea of 2x learning. You know, you can learn twice as much, you can learn twice as much. And even our own families were like, we don’t, we don’t care. Like, why does my kid need to learn twice as much? It’s not a big deal. And we, we’d have like, parent conferences where we’d be saying, hey, if, if your son, you know, hits his, his goals, he can be learning twice as much. And they didn’t care. And then we had this unlock idea of let’s call it two hour learning and say, hey, if your son hits his goals, he can be out of here in two hours and freed up to go do the rest of the things, you know, that he wants to do during the day. And suddenly the parents are like, Johnny, come on, get with it.

Let’s hit our goals. And it was that mind shift of, you know, let’s get your academics done in two hours. And as a side note, you’ll learn twice as much, but let’s do that for two hours. And then one of the code names we actually had for our learning platform was 鈥淭ime Back.鈥� And we went through a whole process in the last year trying to make sure, what’s our new name going to be? What are we going to call this? And ultimately we landed back on exactly what it is that we’re giving kids, which is time back to go do all these other exciting, interesting things during the rest of the day. Because it doesn’t take all day to educate kids. You can not just do academics, but crush academics in a much shorter period of time when you’ve got this personalized mastery-based tutoring.

Transforming Education Models

Well, and I think you’re speaking to, like, there’s many reasons why Alpha has done what many education startups struggle with, which is jumping into the mainstream narrative. And that sense of giving kids back their most precious resource, time is clearly part of it. AI is another part of it. And that’s where we want to dig in with you today, just given the focus of the podcast that we’ve had here. But let me perhaps frame it this way. We now have two school founders on this show, you and Diane, who have each created models that at one level I think look awfully similar in certain respects. If you mix in, say, Rocketship Education or something like that, which was founded in 2006 and is an elementary school model.

Michael Horn: We can take that and Summit Public Schools that Diane founded and Rocketship and say, hey, a lot of the structures that Alpha Schools has at one level, like a relatively limited block of time on learning academics and content in ways that are personalized for the learners, large blocks of time for projects, a big focus on skill development and habits of success or life skills like growth, mindset, agency, and so forth, those are things that were present in models like that. But then we come to at least one big difference, which is, yes, Alpha was originally designed, as you said, right before the mainstream use of AI, just like Summit and Rocketship were. But Alpha is now aggressively developing AI powered dashboards, AI powered learning applications, AI powered knowledge interest, working memory graphs for students. And so, given our focus on the podcast in this particular season around AI, I just love to dive into the AI parts of the model with you. Even as we’ll say up front, like AI is clearly inextricably linked to the other elements of the overall Alpha model. Pulling them apart is not fair to you all. But just given that we’ve heard so many podcasts with you about Alpha, and we suspect most of our particular listeners have as well, I think digging into that AI question in particular, and this is maybe the framing we can bring to it, which is, what does AI allow us to do today? That was not possible in the best of the personalized models from a decade or two earlier.

MacKenzie Price: Yeah, I think that’s a great way to frame it, because artificial intelligence in the learning science world now is what I believe is like the microscope to biology. It is the tool that is finally enabling us to integrate all of these learning science principles that have been known for many, many years can result in kids learning 2, 5, 10 times faster. It just was never possible to incorporate in obviously in a teacher in front of the classroom model, but even more importantly, even in an individualized adaptive app type setting. And so to give context to that, you know, when we first started our school back in 2014, we knew that we could use apps. So we were using things like Dreambox and Khan Academy and Freckle and Grammarly and Egump, a lot of the apps that were kind of out there. The difference was it was still hard to manage the way that kids worked through the apps. And so one of the things we found is that there’s a lot of what we call anti-patterns that kids will do when they’re using apps. It could be things like topic shopping.

You know, they jump in and say, hey, I’m going to go to, you know, I’m a fourth grader, but I’m going to try some fifth grade material just because it’s kind of interesting. Oops, it got hard. I’m going to back out of that. I’m going to jump into some third grade material or I’m going to kind of mess around on this or even more just not engaging with the apps. You know, you could have everything from a kid not even sitting in front of his computer or picking his nose or, you know, just rushing through the explanation and not reading it. And that’s where a lot of the big difference is. One thing to kind of just be clear about, we do not use a chatbot in our education platform. Chatbots in education are cheat bots.

And it was interesting. I actually had a big event last week in Austin. The National Governors Association came and toured and we’re learning all about our schools. And I made that comment, you know, we do not use chatbots. They’re cheat bots. 90% of kids are going to use them to cheat. And a couple hours later, there was another vendor who’s basically built a chatbot for education that was like, well, you know, I put him in a, put him in a little bit of an uncomfortable situation. But I think that’s really important to know.

And one of the things I really don’t want to see in our education system is we slap a GPT on every kid’s computer and suddenly say we’re an AI first classroom. Right? And I was actually talking to a Stanford professor a few weeks ago who said, you know, here’s the problem that we’re seeing. Educators are using, you know, chat features, ChatGPT to create lesson plans, you know, and do these things. Kids are using ChatGPTs to write their stuff. Professors or teachers are using ChatGPTs to grade it. And so basically the AI is just talking to each other. Right. And we’ve taken the human out of it and that is totally not what we’re doing.

So there’s kind of two features that I can go into around how we’re using AI in our model.

Diane Tavenner: Yeah, let’s take this piece by piece. MacKenzie, that will be that context is super helpful. Let’s start in the morning block where you’ve already gone a little bit with some of the apps and whatnot. You all roughly have about three hours where students are doing sort of two hours of head down learning that quote academics my language for that is content knowledge. So forgive me if I slip up and use different lingo. And as I understand it, and as you were just sharing, you’re using these apps or adaptive learning products and you named several for us there. But there are some places where you are using apps that, as we understand it, you’ve built for yourself. And this tracks with my summit experience.

Our first choice was always to buy quality products. Second choice was to partner with startups or companies that wanted to work with power users. And last choice was to build our own when it didn’t exist. So I’d love to unpack. Where is it that you’ve determined there wasn’t something good enough and that you have literally built your own application and are using it right now? And are those AI native applications?

AI-Powered Personalized Learning Systems

MacKenzie Price: So we’ve definitely had a number of years to test out a lot of different apps, see what worked well, what didn’t work, where there are gaps. And what I would say is we’ve curated over this period of time which apps are best for which grade levels in which subjects. Not all apps are created equal, but to kind of start at the very beginning where we’re using AI, we are using AI to be able to assess what a student knows and what they don’t know. So any student who comes into our Alpha school to start takes an NWEA math assessment. We also do math assessments three times a year for all students and that’s how we’re measuring growth. But what we do is we take the information that comes through that assessment as well as some other initial assessments that we’re able to do with students. And from there we have AI tools that will basically build out the personalized lesson plans that say, all right, here’s where a kid needs to go, here’s how we whole fill, which of course is a very common issue. Even our students who come into us with, you know, A’s on their transcripts, you know, can be three years behind in academic content.

Right. Actually we found out students who came in to us this year from other schools, if they had a B on their transcript, they were between three years behind and seven years behind. Which actually shows, you know, grades mean nothing anymore in this day and age. So we take the assessment and we have an AI tool that basically builds that out. So what does that look like?

Diane Tavenner: And that’s a tool you all have built internally, is that Time Back?

MacKenzie Price: That’s a tool that we built out. We have built that tool out and that is using standardized third party assessments like Max.

Diane Tavenner: Yeah, the results. And you’re ingesting the results on that.

MacKenzie Price: Exactly. So they build that. So the experience for a student, a student sits down in the morning during their core block of academics and they will log into a dashboard. We have a time back dashboard that a student logs into and says, okay, it’s time to do math. Now in some of our classrooms, kids get a choice of what subject they want to take on first. Other of our classrooms, you know, we have a set thing. Okay, we’re doing math first, then we do reading, you know, then we do language.

Diane Tavenner: And is that based on age?

MacKenzie Price: Depends on the age. Yeah. And, and so it’s, it’s always interesting. You know, what we’re really working on creating is self driven learners who understand their skill of learning to learn. So like if you talk to some of our fourth and fifth graders, you’ll hear some of them say, hey, I usually will choose to take on my hardest subject first when I’m fresh and I’m ready. Right in our kindergarten and first grade classrooms, you know, that’s more, okay, it’s math time, it’s reading time, you know, and it’s kind of subscribed there. But basically what will happen is a student will go into the dashboard, click on the subject that they are going to take on. So that’s math as an example.

And then the dashboard takes them to the app that has been determined is the one that is right for them and what they’re doing. Now when I say right for them, we also as a school have kind of used certain things. For example, Math Academy is a third party app that we love. We think Math Academy is amazing. They’ve been fantastic partners to work with and it works really great for basically third through high school. We were Using another app for our younger students, earlier this fall, we were using Synthesis, which, you know, that’s a sexy app that, you know, parents kind of like, because kids are doing interesting things. We were seeing, though, like, I don’t know if we’re getting the results we want.

So we’ve made changes, you know, to that, but they’ll go to the level that they need. So you’ve got a fifth grader who maybe needs to go back and revisit concepts from third grade. You know, they have to hit this fast math, you know, concept, or they’re looking at these fractions or whatever it is. So it takes them to that lesson and they’re doing that. So that’s the first use of AI that we have. Now the second use that we use is the vision model. So what’s happening is we’re using an AI tool that we have built that tracks the screen and is actually watching to understand how is a student moving through this material.

So, for example, when they are doing reading comprehension, are they rushing through the article? Are they just scrolling to the bottom of the screen and randomly guessing, or are they taking the time? And of course, you can tell this is a reading article that normally would take, you know, 69 seconds to read. And this kid just answered it within 10 seconds. Okay, now we’re realizing we’re. We have an anti pattern, which is basically an improper use of engaging with the apps. So we’re looking at that in terms of the vision model to see how kids are learning. When they get a question wrong, are they watching the video? Are they, you know, taking time to read the explanation? And then our AI tutor creates coaching for that student.

So it’ll say, hey, buddy, we’re realizing that, you know, you’re not reading the explanation when you get a question wrong. If you take this time to go forward, here’s what it would do. And so we’re basically giving coaching. Now. The other thing is, in our schools, we also have our cameras turned on and they are recording the students. So they’re seeing if you know, the.

Monitoring and Progress Tracking

MacKenzie Price: If the computer has been, you know, quiet for a minute and a half, is it because the student’s not even in front of their computer, or is it because they’re goofing around with their buddy next to them, what is it that they’re doing? And so it’s able to do that. Now our families have the ability to turn that feature off at home if their students are using that feature at home or if they’re working at home, they can turn that off. But in our schools, we do require that that be turned on. And so we’re able to kind of look at the coaching. Now students will basically walk through each of their core subjects, generally in about 25 minute Pomodoro sessions, and then they’re done with their academics in that two hours. The other feature that we’re using with our AI tool is we can really well analyze and understand how a kid is progressing through the material. You know, what percentage completion are they on each of the different apps, you know, and grade level subjects, things like that.

How many minutes do we anticipate? How many weeks will it take before they’re finished with, you know, fifth-grade math? If they put an hour of homework in a night, here’s how much shorter that will take. And one of the things that people love about that, not only do our students get to really see and understand, they have a sense of ownership over their academic journey. But of course, parents can log in, you know, every day if they want to, to be able to see what is my kid working on. What, you know, did he hit his goals? And then what. What we’re also tracking in the way that goal setting works is students are getting experience points, XP, to borrow, you know, a term from video gaming. And so the goal is that they get 120 XPs per day, which is 120 minutes of focused work. That’s one XP is equal to one minute of focus work.

And so that’s what we’re working on. And then when you ask about the apps that we’re using, we have built Alpha Math, Alpha Read and Alpha Write are some of the apps that we’ve incorporated into our model. And then we’ve got some other things that, you know, that we’re continuing to roll out. One that’s actually available to the public for free is an app that we’ve built that helps encourage the love of reading, which of course is a difference between learning to read and learning to love to read. And that’s called teachtales.com and you can go to teachtales.com and basically it’s using AI to generate personalized reading material based on a student’s interests that then delivers at the appropriate Lexile level for them.

Diane Tavenner: Awesome. There was a lot in there. So let’s.

MacKenzie Price: There was a lot. I need to work on more short sound bites. Well, I hope that doesn’t get worse as I get older.

Diane Tavenner: We all have things we need to work on, right? Let’s stick with those three apps that you’ve developed. So Alpha math, read and write. Are you using those across all of your grade levels? And are they AI, are they adapt, are they AI native, are they adaptive? What’s going on with those apps?

MacKenzie Price: So the Alpha Write is something that we’ve been really excited about and we break this down just to have an idea of how the app works. We break this down with the idea of can you write a grammatically correct sentence, you know, then building onto paragraphs, then building on to essays and working through. And I will tell you, I mean, we had a lot of students, again, A students from their previous schools that come into Alpha. We had high school students who couldn’t write third grade level sentences, like, it’s just crazy how poorly this is going.

Diane Tavenner: Yeah, that’s one of the questions I think that comes up is where writing is situated in the model. So it sounds like you’ve got writing in the morning block as sort of a standalone kind of just expository approach to writing.

MacKenzie Price: We do have writing in the morning block now. Our students are also doing a lot of writing in the afternoon. So, you know, for example, they’re writing, you know, talks that they’re going to give for TED talks, they’re writing essays, they’re writing book reflections that are part of our afternoon block, which is our check chart time. So it is a common fallacy that people have of, oh, these students aren’t actually doing a lot of writing. They’re absolutely getting, they get a lot of writing in. But we’re really breaking this down into everything we’re kind of thinking about is what actually works when it comes to educating students. And where have we been doing it wrong? And that’s where I think it’s so exciting to see all these learning science principles that can come up. And you know, for example, here’s another thing that we do during these, the, the core block period.

Optimizing Learning

MacKenzie Price: We’re, we’re measuring what percentage accuracy students are at to understand are they in the zone of proximal development. Right. If they’re getting more than 85% of the questions right, you know, then that’s a sign that they’re, they’re in too easy material. If, you know, they’re under 70, it’s a sign this is too hard. How do you make sure that they’re staying in the right spot? And so that’s the other part that the AI tool will kind of say, whoa, hold on here. We’re noticing that there’s something changing or that a student’s not being hit at that right level. The other thing that’s going to come in to play is we’re also going to be able to really take a lot of things around cognitive load theory principles and understand, OK, if a student only needs 5 reps of a concept in order to master that concept, they shouldn’t have to sit around and do 10 reps. And if the student needs 15, they shouldn’t only get 10.

So that’s just some ideas of some of the things that are coming in the pipeline that generative AI is going to make really available.

Diane Tavenner: So two things I’m trying to understand and contrast to pre AI to now that we have AI because a lot of what you’re describing sounds very much like what Summit Learning was about. You know, we built thousands of playlists and young people, they actually had a lot of choices. So we were working on self direction in, you know, they would do a pre assessment, they would know what they know, they would prepare, you know, and study and learn. And then they would take a post assessment, we would assess all the things you’re talking about. So I guess I’m wondering in these apps, is that similar or is AI actually playing a new and different role here? And then I do want to get to the sort of time back coach as well because I realize it’s connected. But, are we using AI in these apps? Are these sort of still adaptive learning apps? Are they 鈥�?

MacKenzie Price: Yeah, the third party apps that we’re using are not using, you know, an AI feature and they’re not creating dynamic content. You know that, that is created. This is, you know, The K-8 Common Core curriculum is what’s, what’s being fed into these apps. Where we are getting to is we are going to be moving in, in 26 to dynamically created content. Obviously there’s been a problem. There’s still hallucination issues. In fact, we have a group of high school students, kind of our, our top honors students who we are testing out dynamic content and they’re able to say, hey, guess what? The AI is acting up here. Like this is totally a wrong question on that.

But right now what we’re doing is we’re going through and we’re analyzing every lesson before it’s out there. So this isn’t just like an LLM creating a fifth grade curriculum. We’re still using that. Where the AI tool is really being used is around that vision model. So that’s the biggest difference is that, and that’s part of the reason, you know, if you talk to families who went to Alpha, you know, six years ago, you’ll hear a much more varied experience. Right. We had a lot of families that my kid wasn’t learning.

They were goofing around. There wasn’t this connection. Now there were a lot of reasons for that. We didn’t have the motivation model locked in. We didn’t have the high standards, just expectation. But the other big part was it’s really easy to goof around when you’re learning on these, you know, in general on these apps. And so that’s the biggest thing right now is that our AI tutor is ensuring that kids are moving efficiently at the right level and then understanding what the pace is for that and creating basically new lessons that will fill academic holes, you know, and go at their pace, is what I would say. But yeah, if you’re looking at, you know, for example, a math academy, you know, type of thing, you know, that is static content that, that kids move through and kind of work on.

We used to use IXL, actually. IXL kicked us off of their platform. They don’t like us for some reason. They literally won’t even tell us, they won’t talk to us. They just say, you’re off. But we had used IXL a lot. And actually one of the things I always say for families that are wanting to recreate this at home, I actually think IXL does a really good job across a lot of dimensions. They were a pretty good app.

They don’t like Alpha for whatever reason, but, you know, that’s where we’ve kind of been able to figure out what this is. But I think the other question is, when you talk about things like reading, writing, it’s really helping break down our apps that we built. You know, they’re breaking down into small components. Let’s make sure a student is excellent at this and then build from there. I think in a traditional classroom, having students write a five paragraph essay is not necessarily helpful. Instead, are they really understanding the structure and mechanics of a sentence? Are they understanding what a paragraph should look like? Are they going. And we use really the idea of building blocks in all of the work that we do.

Diane Tavenner: So does that mean you’ve got under underlying at least the apps you’re building sort of a knowledge graph that you’re, that you’re working with in order? Yeah, I mean that again, fairly. Okay, fairly consistent. Let’s dig into that AI coach or tutor, like you said, because it sounds like this is not a traditional dashboard where young people are looking at Their own data and information. Maybe they are. But what it sounds like you’ve really got is this AI coach or tutor coming in to keep them motivated. I mean, the apps you’re talking about, lots of schools have them, as, you know, lots of schools, they just don’t get the number of minutes, they don’t get the progress. And so is you’re. It sounds like that’s the key.

So that is an AI tutor or. But it’s not a bot that you were referencing.

MacKenzie Price: Well, it is, but you’re not correct about. Yeah, you’re not correct about that. The AI tutor is not providing the motivation levers. There’s no motivation that’s happening through the apps. The motivation is all through our guides, our human teachers. They are focused on motivation. And just to be really clear, the reason we’re having the success that we’re having and the academic results we’re having is not because of our ed tech. Our ed tech is fine, it’s whatever.

But there is no magical edtech product that just immediately motivates and makes a guide or makes a student, you know, lock in and be able to learn well, We haven’t built it. We haven’t seen it yet. The key for us is that we have freed up the time of our human adults to be able to focus on motivation. And so that could be everything from, well, from the idea that students earn alpha bucks for hitting their XP goals to, I was just talking to one of our kindergarten guides the other day, and she said, you know, we have kids where when they hit, one of their goals, when they. When they unlock a goal that they.

They’ve done, they have a secret sniggle, they have a secret signal, they’ll, you know, scratch their nose. And that signals, oh, you hit a goal, let’s do a silent dance party. And It’ll be a 15 second, you know, the guide is doing the silent dance party, and then they move on to the next thing. It can be individual motivation, you know, models. We had a student who, as a result of hitting her academic goals over a period of six weeks, she earned time in a professional recording studio to record an original song that she had written and was singing. So that’s the whole key. And by the way, 90% of what creates a great learner is a motivated student.

10% is having the right level and pace, which is what our edtech tool does. What the AI tutor does, though, it actually does give kids the ability to go on their dashboard and each day and see, okay, I hit my rings, I filled my ring. It kind of looks almost think of an Apple watch, you know, with exercise rings. That’s what it is for each student is, did you fill your ring? Which means, did you get your XPs in that subject? And then they can go into their learning dashboard and they can see at any time, here’s how much I. Here’s how much I hit. We even have a waste meter in the corner that says, you know, you’ve wasted 20% of your time you were wasting by not engaging in the right way or not accurately doing that.

Diane Tavenner: So the student doesn’t actually, like, engage with the AI tutor. It literally is just powering this dashboard then.

MacKenzie Price: Well, it’s powering the dashboard, and then it will pop up and say, you know, it’ll write something like, hey, watch the video explanation. You know, sometimes it’s, you know, going.

Diane Tavenner: It was like a nudge or something.

MacKenzie Price: One of the things that, yeah, we’ll see is that, you know, we’ll often say to students, you know, often the fastest way forward is to slow down, slow down and read the explanation. So it does that. But here’s what it’s not doing. There’s not some little avatar Dashy, that pops up and is like, hey, Johnny, you’re doing such a great job. Two more questions, and then we’re doing that. It’s not that kind of thing. The AI really is kind of under cover.

And it’s again, building these lesson plans and then analyzing and understanding how a kid is moving through that.

Diane Tavenner: Building the lesson plans that are in the apps or in the …

MacKenzie Price: Yeah, taking them to the right spot. So it’s able to say, okay, we’re going to take you.

Diane Tavenner: Oh, by lesson plan, you’re saying directing them to specific.

MacKenzie Price: Directing directly to this math academy. And we put up these basically guardrails. That don’t allow a kid to pop out of Math Academy and say, hey, instead of doing this concept, I’m going to go play over here. I’m going to go do this. And I think that’s a problem in traditional classrooms when people are using apps. They’re given their iPad or their Chromebook, they’re put on Khan Academy, and then they’ve got the ability to kind of bounce around. There’s one other topic that I think is also important, and this is actually a lesson we learned very early on, is the idea of requiring students to do some work each day in each subject. Right.

And there’s a lot of alternative education systems that’ll say, hey, if a kid doesn’t really want to focus on math for a couple months, that’s okay. They want to pursue reading. We actually believe. And this was, I’ll never forget the very first year we had a first grade student who absolutely loved math. Loved math. He was at 8th grade level math. And the problem was he needed his guide to read the word problems to him because he couldn’t read and he hadn’t read in like months. And that was one of the early unlocks where we realized, okay, we have to require, you know, time in each subject each day that students are accomplishing, which some, again, some alternative schools don’t do that.

Diane Tavenner: Yeah. So it sounds like then, the motivation is highly related to this relationship that young people have, which we know is very powerful. And then just following the directives essentially of the guide and then the technology to do what you’re telling them to do and stay on track.

Confidence Unlocks Student Motivation

MacKenzie Price: Exactly. And then I think the next part of the motivation, kind of the deeper level of motivation is and you know, people often go, oh, is extrinsic motivation bad? And you guys know, there’s all the research that shows there’s not necessarily even that same, you know, intrinsic versus extrinsic. But what we are seeing is that as students become more and more capable, you know, and build up their knowledge, they become more confident and they do get more motivated. They suddenly realize like, wow, okay, I can be 99th percentile in, you know, math, in language, in science, I can do this, it’s not as hard. And so we find that kids, their identity really changes as they start to see that, wow, I’m capable of learning when I’m given the right level and the right pacing and I get motivated to do that. And that is what I think is the really cool unlock that we enjoy seeing when students finally realize this. Like, wow, I can do this.

Diane Tavenner: Yeah, definitely. You said that one of the benefits of this approach is you freeing up the guide time to really do the more important things. And as I understand it, one of those activities they do is one to one meetings with the young people in this morning block. This was one of them. Continues to be, I think the most highly rated element of the summit model is the mentoring model with the one to one check ins as a part of that. And over the years we started leveraging technology to enhance those check ins. I’m curious if you’re using AI in any way to support the one to one check ins and, and what that looks like.

MacKenzie Price: Yes, we are. So we actually mic up the guides during those one to one check ins and then they’re using, you know, we take those transcripts and we’re running them through for everything from what percentage of the time were you talking compared to the student? Right. If you’re talking too much, that’s a problem. How many questions were you asking, you know, versus stating what are some of the things that are happening there. We also actually use that technology for some of our students as well. So an example of that, one of our students in Arizona, he struggles with a growth mindset, you know and he’ll, when he’s struggling in his academic work, he’s quick to say I’m dumb or I can’t do this or whatever. And so we put an AI mic on him and then he and his guide go through daily and analyze how are you speaking to yourself? Were you being kind to yourself? And what we found amazingly is that just him knowing he has this lanyard around his neck that’s listening helps him remember, hey, speak kindly to myself. I can incorporate these growth mindset strategies.

So we’re able to do that. We have guides that wear these lanyards throughout the entire day so that they can understand and then get feedback on their coaching. And so, you know, that’s, that’s a great part of it. We’re using AI. We’re very much, our organization is very much on be AI first in everything we do. How can we always take everything to the next level and build that out? And then of course the other aspect of AI, you know, that comes across in our afternoon life skills workshops is kids are learning how to use these tools that are going to help them be successful. So you know, kids are starting to build out and develop these brainless and then build out an LLM. In fact, we actually just had a pretty exciting thing happen last week.

One of our students at our high school had built up an LLM around safe teen dating advice and she ran a research study with the University of Texas professor around basically how good was the LLM she built compared to a ChatGPT and suburban moms and they just submitted to Nature with that research information. So it’ll be really exciting in the next couple of months. We’ll hear if that gets accepted. And that should be a pretty cool thing. So that’s the other part of this is you’ve got to make sure kids are being equipped to learn how to take advantage of all these new tools that are constantly coming out.

Diane Tavenner: For sure, for sure. Let’s move to that afternoon block and unpack that a little because I think I hear far less about the afternoon time, which is familiar to me, because also in the Summit model, you know, the self directed learning time seemed to get all of the publicity in the play and whatnot. It was only two hours. It was only 30% of the young person’s grade, but it got like 90% of the attention. So let’s break the afternoon into the K8 and the high school because I think those two are different in your model. Talk about the K8. Yeah, talk about the K8 afternoon, where I understand it’s young people are learning life skills. Is this a project based approach? Who’s planning this? Is it a curriculum? I think, as you just said, students are encouraged to use AI from their side.

But what I’m really interested in is how are guides and educators using technology and specifically AI for this afternoon block, the dashboard here. What’s going on there?

MacKenzie Price: Yeah, this afternoon block is really when our guides are shining in terms of being able to plan and connect and mentor our students. And that’s done a few different ways. So when we think about In K through 8, our students are participating in these life skills workshops that are developing leadership and teamwork, financial literacy and entrepreneurship, relationship building and socialization, public speaking and storytelling and grit and hard work. And so every workshop that is created has to be able to pass two tests. One is, what is the life skill that is actually being taught and how are we going to assess at the end of the six week period whether that has happened? So, for example, you know, we’re in the week before the holiday break. We’ve got test to pass events happening at all of our schools around the country where parents and people from the public can come in and see something that’s being done that the kids have been working on and understanding. Did they learn this life skill? You know, an example that we often talk about because I think it really highlights the idea of how do you learn grit? How do you learn, you know, stick with something when it’s hard? So we have students who participate in grit triathlons. And that could be things like having to solve a Rubik’s Cube, juggling three items for 30 seconds and running a mile without stopping.

And when you can see that a kid has, you know, a third grade student has been able to understand, okay, there’s an algorithm and I keep practicing my Rubik’s Cube and I start by juggling scarves and eventually I’m juggling balls and I incorporate atomic habits to, you know, walk and run. At the end of six weeks when these students are able to accomplish that goal. And it shows grit. We also do a lot of physical workshops that build out things like grit, like facing fears. For example, we’ve got a rock climbing workshop and that actually for our kindergarteners, they’re climbing a 40 foot rock wall. And when you watch the difference between a student at the beginning of that six week period, you’ve got a five year old who’s like, I don’t even think I can hold on to one of these suddenly going 40ft up. The only one more amazed by that are their parents, right? Their parents are like, this is amazing. So a lot of physical workshops that are doing, doing things and then the guides will use AI tools as part of building out those workshops. Being able to measure one workshop that we do every year that’s very popular.

It’s a communication and basically uplifting others workshop. And the test to pass for that workshop is that kids go into an escape room, you know, one of these, one of these rooms where they have to, you know, solve a bunch of different puzzles and logic things and all that to go. And we mic the students up and we use AI to analyze what percentage of their language is considered uplifting and positive. You know, where are they doing that? We’ll do that in sports activities. Kids will get feedback on their public speaking. They’ll be using AI tools to build graphic novels, to build films, you know, all kinds of things that they’re working on that way. And so that’s a combination of group workshops. And then they also get individual time to pursue what we call kind of check chart independent projects.

Diane Tavenner: Ah, so it sounds like then your guides are using just AI, like an LLM to help them plan those workshops. And then are you rubric gradient or just checklist grading?

MacKenzie Price: We’re rubric grading as well. And so we have for each life skills workshop we’re grading, what is the quality of workshop. And that’s everything from, you know, the kids’ assessment of did they love the workshop. You know, we’re constantly surveying parents, kids to make sure that what we’re delivering is right. And how are these guys going? The thing that we’re calling it.

Diane Tavenner: And that feedback from the rubric, is that derived from the AI or is the guide doing that? And then is that also incorporated in their dashboard?

Iterating to Build Measurable Skills

MacKenzie Price: All a combination of both things. And I think in a lot of ways what we are constantly doing is iterating. How do we build upon a workshop, how do we make, are we doing each session that kind of comes together. In fact, you know, today again, it’s the last week before the holiday break. We’ve got staff days every evening, you know, after school as we kind of plan and go through what worked, what are we doing to kind of increase, you know, love of school, the learning 2x in 2 hours and then development of life skills. So we’re working through a lot of these types of activities of, you know, how can we make this alpha life core soft skills measurable? Right. How can we understand how to measure these skills versus just kind of saying oh, you know, sure, they’re learning leadership qualities, you know, from, from something. What are the things that we can do to, to kind of build that out?

Diane Tavenner: Interesting. One of the conversations, big conversations, is how AI can and should change the role of the educator. And you all have purposely and publicly redefined the role of the teacher to be a guide. And I’ve been tracking through this conversation. You know what I think some of the shifts are in how you think about teacher versus guide and educator and how AI is enabling that. So let me run this back past by you and see if I got it right. So the guide’s not planning any sort of lectures or traditional lessons and they’re not doing any assessment. They’re leaving that to the technology.

They are doing one to one check ins and they’re getting feedback from sort of AI inputs from their recordings and things like that about how they can improve. So that takes time. We know in a teacher’s day if you’re transcripting all of those things, they’re going to an educator’s day and then they are planning the afternoon workshops. It does sound like they’re doing some of the assessment there. And they’re certainly, you know, working closely with the students on the motivation piece and engaging directly with them. And it does sound like that’s supplemented by AI. Did I get that right? Sort of the role of the guide, if you will.

MacKenzie Price: Yeah, you did get that right. Now there’s one other aspect of the guide’s job, in the morning academic time, in the core time. You know, I think people have this, this misconception that oh, you know, you’ve got a kid, a group of kids that are just staring at computers with no adults in sight. Our guides are there and they’re engaged, but they’re not there to teach academics. So if a kid says, hey, I’m struggling with this, you’re not going to see one of our guides saying, okay, let me, let me show you how to work through this problem. You got to carry the one. Let’s do a tutoring session on this. Instead.

They’re going to be basically asking students questions to help them understand if they have used their resources. So, hey, were you able to watch the video? Did you go into the resource library to find another answer? Did you check these kinds of things out? And so that’s where they’re really providing coaching around how to go about learning to learn. Here’s one. I don’t know if you call it an exception, but one thing I will say for our younger students, our kindergarten, first and second, we have not found to this point a replacement for reading than that one to one reading time. So we have reading specialists at all of our schools for our younger learners who are working with students on reading. And our students get one to one pull out time, you know, to be practicing that reading. It’s something critical. We are seeing, you know, certainly some great progress and success around learning to read.

But you know, you have to have that time reading out loud with a human. And so that’s the one thing I would say is our guides in our younger levels, we do have certified like reading specialists who are at those schools. And it’s, it’s critical.

Diane Tavenner: We didn’t talk about the high school afternoon time. And as I think you alluded to, and as I understand it, this is where young people are picking one project to work on for four years. And again, I don’t know if that’s a headline or if that’s accurate. I must say this is an element of the model that gives me a little bit of pause and so I’d really love to underbutt a lot of buzz. So what’s actually happening for high school students for those four hours, four years?

MacKenzie Price: You know, so we have two tracks for our high school. We have what we call an honors track. And the idea of that honors track is basically kids who kind of, you know, want to be sort of Ivy League bound. They’ve got ambitions of going into a top 20 university. And so in that program we’re basically saying, okay, we’ll deliver 1550 SAT score scores, you know, fives on at least a few hard AP courses and what we call an Olympic level Alpha X project. This is a project that is as impressive as being an Olympian. You know, what is it? So an example of that, one of our students who just got accepted to Stanford this past week. She’s the student who’s also submitting her research to Nature.

If she’s accepted, she’ll be the youngest female ever and the only high school student in history. You know, to be able to do that, you know, they work on something big. Now during that time when they’re working on these Alpha X projects, there’s no question that you’ll have kids who might, they might decide to change their project 10 times during their four year experience. What they’re really developing is the skill of learning how to go deep into something and become an expert. And so we’ll do things like they’ll go into, you know, two week long sprints where it’s like, go learn everything you can learn about this subject. And at the end of that two weeks, you know, just as often as not, you’ll have kids come out and go, actually it turns out I’m not interested in that. I want to go into something else. And the other thing is these projects that kids work on aren’t necessarily what they say, oh, I’m going to do this for the rest of my life.

Right. I’m going to go build this out in college or something. But it’s a project that they’re kind of, you know, able to develop and go deep and become an expert on. Now we also have a non honors track at our school and that non honors track is for kids who say, you know, I really love the idea of getting time back to just go do things I’m interested in. So for example, you know, we’ve got a student who wants to get his pilot’s license and he loves the idea of flying planes. Now does having your pilot’s license at age 15 get you into Stanford? Yeah, you know, maybe not, but it gives you time to go develop these things. So a lot of our athletes who want to have time to pursue their sports or whatever. Now what all of our students do, and that non honors program basically is 1350 SAT, which is, you know, top 10%, fours and fives on APs, you know, and time to go and develop the interests that they have. Honors students are spending about three hours a day on their core learning.

The non honors track is about two hours of what they’re doing. Kids are still taking AP courses, they’re still doing all those kinds of things.

Diane Tavenner: Sorry, you lost me for a second. Where’s the AP course? Is that in the afternoon or in.

MacKenzie Price: No, that’s in the morning. That’s the core academic time is students are taking four years of English, four years of math language or, you know, foreign language, all that kind of stuff. So they’re doing that in the morning. Afternoons are for working on these Alpha X projects. And then we do a lot of workshops around life skills for all of our students. So that’s everything from rejection training to giving and receiving feedback, you know, leadership challenges. A lot of things that students are working to kind of build out those skills is what our high school program looks like.

Diane Tavenner: So in the high school afternoon, there is sort of still a framework curriculum. Maybe it’s not every day, all the days, but that you do have some of these skills that you’re doing in some workshop, being around with students.

Developing Projects with Real Impact

MacKenzie Price: Yeah, there’s absolutely a framework. And then for the kids who are working on their Alpha X projects, they basically go through different levels, right? So, you know, as an example of the kind of the highest level where basically these kids are getting out and they’re launching real businesses or activities. One of our students, who’s the senior this year, she’s working on getting a musical launched on Broadway. So she actually spends, you know, five, five to seven days a month in New York City, you know, working on recording with producers, meeting with potential investors, you know, doing those types of activities. So she’s kind of been released out into the wild, you know, in some ways to go work on these projects. But the other thing that we have in common is every day our students are spending an hour working on their brain lift. So this idea of whatever the interest they have, they’re staying current on research, what’s going on, and they’re using this brain lift to then build out whatever their LLM and GPT is based on this. They also work on things like creating a spiky point of view.

So an example of that, we have a student named Alex who is building a plushie doll that is basically a mental health coach. And his spiky point of view that he’s built is he believes AI can actually provide better counseling to a teenager than a human counselor. Now, that’s a very spiky point of view, right? Especially when you think of all of the dangers on this. But he’s built certain things in his system that he believes are making a successful AI mental health coach. And so the idea is building out these things and being able to learn how to become an expert on using AI to build this thing out. So we have another student who’s interested in creating. He’s a filmmaker and wants to create, you know, his ultimate goal is to create an Oscar winning, winning film.

And part of what he’s done is to create basically a spiky point of view around how filmmaking can be done. And he just got accepted. He reached out to a bunch of different podcasts. He got accepted and invited on three podcasts. Now a lot of rejection training going on in there as well, where there’s a lot of podcasts who say no answer, you know, or whatever it is they do. But they’re learning all of these skills during this time. Plus getting the traditional academics that, you know, students in a normal school are getting.

Diane Tavenner: Where would science labs fit into this model? Or, you know, projects that are in history where we know kids, you know, dates, facts, information is, is based, but you actually need to understand the big themes and trends. Where does that fit in your model?

MacKenzie Price: Well, if you take things like science labs. We don’t have science labs. Our students are taking AP Biology, AP Physics, AP Chemistry. But they are, you know, watching great YouTube videos that are exploring these topics instead. We haven’t found that there’s this critical piece of getting kids in a lab doing beaker experiments, you know, as part of what they’re doing. They can watch these things. Now. Kids who are really excited about something that they’re working on, you know, in science can go in and build something out.

So for example, we had a student who got really interested in cancer research and epigenetics, and she ended up going out and creating a documentary that’s been viewed over 5 million times around cancer and epigenetics. So we kind of think like everything we do at these schools is taking an interest or a passion that a kid has and figuring out how to get them out in kind of real world experience with things and how they can build. We had a student who loves physics, really interested in science, loves physics. He also went on to become a professional water skier, but he would take physics principles and then work on how he could improve his water skiing times and rope length, you know, incorporating physics principles. So there’s things they do there, things like history, for example. You know, students are taking AP World and AP European and AP US History. So they’re doing all those things. They’re getting a lot of experience on writing, obviously, as they’re, they’re learning on apps, they’re coming out with, you know, fives on their APs and doing very well, and they’re having some connected time with each other where they’re, they’re basically going through some checkpoints at the same time.

Where they’re interacting last year towards, you know, basically in April you heard a lot of singing because kids had basically used AI tools to help them remember a bunch of their facts for AP world history, you know, with basically in the, in the same vein as Hamilton lyrics, you know, and, and working through those things.

Diane Tavenner: Is that the College Board’s digital curriculum that they’re using for the AP courses? Yeah. And then, that like joint collaborative time would be in the afternoon. Is that how it connects?

MacKenzie Price: Yeah.

Diane Tavenner: Got it. Awesome.

Michael Horn: This season of Class Disrupted is sponsored by Learner Studio, a nonprofit motivated by one question. What will young people need to be inspired and prepared to flourish in the age of AI as individuals, in careers and for civil thriving? Learner Studio is sponsoring this season on AI in Education. Because in this critical moment, we need more than just hype. We need authentic conversations asking the right questions from a place of real curiosity and learning. You can learn more about Learners Studio’s mission and the innovators who inspire them at www.learnerstudio.org.

Michael Horn: This has been super helpful, MacKenzie. Huge thanks. But before we let you go, we have this segment where we, where we get away from the conversation around education generally, although not always. Just things we’ve been reading, watching, listening outside of work if you can. But if not, that’s cool too. So we’ll let you have the first say at it before Diane shares what’s been on her list.

MacKenzie Price: Well, I’m sure that I’m going to give you an answer that is not going to be impressive to any of your followers or listeners.

Michael Horn: I guarantee you most of my answers are unimpressive. So go ahead.

MacKenzie Price: My absolute favorite thing to do in the evening when I get time to relax is I love to take a bath and I have a huge television that is mounted in my bathroom in front of my bathtub that is non-negotiable. My husband and I just moved into an apartment a year ago and I was like where is the TV in front of the bathtub going to go? Like I will not move into an apartment that doesn’t have that option. And I got in the bath last night and I was so excited to watch the Taylor Swift Eras documentary. So I am halfway through the first episode. My girls and I, and actually my husband too, we totally bond over that. And then actually later in the evening my daughter’s home from college and we’re watching this show called All Her Fault. It’s like about a kidnapping and it’s the gal from Succession, you know, the redhead from Succession, she stars in it. And one of the guys from White Lotus season one.

So I do. We like those types of shows. We loved White Lotus. This All Her Fault. I just watched the Beast in Me. So I do, I sometimes can be known to binge some of these Netflix shows, but I do them in the format of about 35 minutes, which is how long my bathtub water stays hot for. And then I’m out of time.

Michael Horn: And then you’re out.

Diane Tavenner: There you go. Well, I’m totally, I’m totally cheating today. I’m gonna share a novel that I’m going to read over the holidays. My favorite living authors, Ian McEwan. And he has a newish novel out called What We Can Know. And I, I’m literally counting down the days to the holidays and to being able to crack this one open and savor it. I’ll give you two sentences from the New York Times review that make me excited. Quote, it’s a piece of late career showmanship.

McEwan is 77 from an old master. It gave me so much pleasure, I sometimes felt like laughing. I will report back.

Michael Horn: And you’ll have to report back because I was going to say you just quoted the New York Times, which is an item for later but yeah, so, all right, I’ll wrap with mine, which is MacKenzie, to your point. We binge watched Four Seasons with Tina Fey and Steve Carell. It’s a Netflix. I hadn’t heard of it. It’s like an eight episode first season. There will be a second season based on the cliffhanger at the end. And I would say it’s about three couples, sort of 50s age group is roughly where they are and through trials and tribulations that is hysterical.

A lot of predictability and yet still very funny as it went through. So we really enjoyed it and I think binge watched it in two nights. I think so.

MacKenzie Price: Oh, great. That might be our holiday activity too for some time.

Michael Horn: There you go adding to your.

MacKenzie Price: I love that. I love that.

Michael Horn: Awesome. Awesome. Well, MacKenzie, huge thanks and as always, huge thank you to you, all of you, for listening. Keep coming with your questions, comments and all the rest, and we’ll see you next time on Class Disrupted.

This episode is sponsored by LearnerStudio.

Keynote speakers this year include Monica J. Sutton, creator and host of the children’s education series Circle Time with Ms. Monica, Yale psychology professor and Happiness Lab podcast host Dr. Laurie Santos, appearing alongside Common Sense Media’s Bruce Reed, and bestselling author Jennifer B. Wallace, whose work centers on the human need to feel valued 鈥� and to add value. 

Also featured: former Presidential Science Advisor Arati Prabhakar, who will join a panel on 鈥渕oonshot鈥� thinking and the future of AI-driven learning. And a new documentary traces the career of longtime Sesame Street star Sonia Manzano.

Artificial intelligence this year plays a bigger role than ever. Dozens of sessions examine AI’s expanding role in classrooms, from adaptive tutoring and authentic assessment to teacher burnout, algorithmic bias and what it means to be literate in an age when machines can write, reason and create.

This year, the Austin Convention Center, which typically hosts the event, is under construction. So sessions will be held at four venues around downtown Austin. Organizers are also planning a 鈥淪XSW EDU Clubhouse鈥� at the historic , which will host daily performances, keynote livestreams and social events each night.

Because of the event鈥檚 multiple venues, space may be limited, so organizers recommend booking reservations for keynotes, featured sessions and workshops. They鈥檝e provided an with details. 

To help guide attendees, we鈥檝e scoured the 2026 to highlight 26 of the most significant presenters, topics and panels:

Monday, March 9:听

9 a.m. 鈥� : Researchers, district leaders and family engagement specialists examine the chronic absenteeism epidemic that has left millions of American students disconnected from school since the COVID pandemic. This panel presents the latest data on what is actually driving absenteeism 鈥� from housing instability and health crises to school climate and whether students feel they matter. It鈥檒l explore which interventions are producing genuine, sustained improvement.

11 a.m. 鈥� : This panel presents evidence that score inflation on standardized tests, state-level proficiency standards and the federal retreat from accountability are making it harder than ever for families to get an accurate picture of their child’s true academic standing 鈥� and what policymakers can do about it.

1:30 p.m. 鈥� : This Opening Keynote features Monica J. Sutton, educator, entrepreneur and creator of Circle Time with Ms. Monica, who traces her journey from preschool classroom to digital learning spaces reaching millions of families worldwide. Sutton challenges educators to evaluate every innovation through a developmental lens, asking: Does this technology honor how young children learn, grow and thrive, while protecting curiosity and connection?

2 p.m. 鈥� : What do real students think about AI? How do they want to learn about it? This session, by MIT Media Lab鈥檚 Jaleesa Trapp and LEGO Education鈥檚 Jenny Nash, explores strategies for building AI literacy through hands-on computer science that fosters critical thinking and ensures safe, responsible AI use.

2 p.m. 鈥� : Civics teachers, researchers and policy advocates will examine how teachers are navigating the nearly impossible task of teaching democracy, elections and civic participation in classrooms where students and families often hold deeply opposed political views. The panel shares new findings from America鈥檚 Promise Alliance鈥檚 State of Young People research and explores strategies for creating classrooms where hard but evidence-based conversations happen productively 鈥� and where students develop the civic skills needed to participate in and repair a fractured democratic system.

4 p.m. 鈥� : Child development experts offer a science-backed framework for evaluating AI for young learners without compromising the play, exploration and human attachment that are foundational to healthy development. This session offers an 鈥渦rgent exploration鈥� of AI’s impact on brain architecture and what educators, parents and policymakers must know to protect young minds.

4 p.m. 鈥� : A panel of educators explores the causes of low student engagement, absenteeism and cheating, sharing classroom-tested solutions for creating assignments that are cheat-resistant by design. Rather than relying on cheat-detection software and pedagogy that punishes students for cheating, panelists will share how to foster a culture of academic integrity based on student agency, purpose and ownership of learning.

4 p.m. 鈥� : In this featured panel, Rep. Jim McGovern (D-Mass.), Chef Ann Foundation CEO Mara Fleishman, University of Pennsylvania student Maya Miller and Duke World Food Policy Center Director Norbert Wilson make an evidence-based case that school nutrition is an educational issue, not merely a logistical one. Panelists connect chronic hunger and poor nutrition directly to cognitive function, attendance, behavior and academic performance, and present district-level models that have transformed school meals into assets for learning.

Tuesday, March 10:

9 a.m. 鈥� : This featured session stars Roya Mahboob, CEO of the Digital Citizen Fund, who will draw on her experience growing up in Afghanistan to trace how exclusion compounds across the pipeline from K鈥�12 classrooms to corporate boardrooms. Mahboob offers evidence-based interventions that have demonstrated real impact on girls’ participation and persistence in tech, as well as a vision for education that is inclusive, practical and full of possibility.

9 a.m. 鈥� : A candid discussion on the science, ethical considerations and implementation challenges of using Voice AI for assessment in K鈥�12 classrooms. Learn what鈥檚 promising, what鈥檚 problematic and what鈥檚 on the horizon as experts explore how Voice AI differs from other AI tools such as large language models (LLMs), and how it can be integrated in ways that truly support students and educators.

12:30 p.m. 鈥� : In this keynote, Bruce Reed, Head of AI at Common Sense Media, and Dr. Laurie Santos, Yale psychology professor and host of The Happiness Lab podcast, examine how rapidly evolving AI technologies and social media are shaping young people’s mental health 鈥� and how families, educators and policymakers can respond. They explore the science of well-being, the risks of algorithm-driven systems and common-sense guardrails to protect young minds. 

2 p.m. 鈥� : This panel challenges the deficit framing that has long defined how schools, families and students themselves understand dyslexia. In an interactive session, a think tank-style panel will present a strength-based model of dyslexia support and examine how AI tools are beginning to unlock academic access for students whose abilities have been systematically undervalued.

3 p.m. 鈥� : Director Anna Toomey’s feature documentary tells the story of five mothers determined to establish the first public school in New York City for children with dyslexia. Toomey follows their battle to open the South Bronx Literacy Academy, addressing a learning disability that affects about 20% of the public. A post-screening discussion connects the film’s themes to national debates about reading instruction and equitable access.

4 p.m. 鈥� : As chronic absenteeism reaches historic highs, schools are doubling down on academics, interventions and incentives. But they may be missing underlying emotional and psychological factors driving absenteeism: stress, anxiety and lack of belonging. This session looks at how rest, youth voice/choice and emotionally safe environments can re-engage students.

5:30 p.m. 鈥� : Director Ernie Bustamante’s feature-length documentary offers a portrait of Sonia Manzano, the trailblazing actress who played Maria on Sesame Street for 44 years. A conversation with Manzano herself follows the screening, exploring how public media can reach children when formal schooling often fails, and what Sesame Street鈥檚 legacy means in the age of AI-generated children’s content.

Wednesday, March 11:听

10 a.m. 鈥� : This performance offers an early look at a show in development that began as a teacher performance at a school meeting. In this Hamilton-meets-The Sound of Music-meets-Good Night and Good Luck story, set against today’s culture wars, three high school students and their teachers navigate questions of identity, purpose and what school can and cannot teach. A Q&A with Peter Nilsson, the show’s creator, follows the performance.

11 a.m. 鈥� : This solo session by Toby Fischer, an Ohio educator, offers a sweeping reimagination of literacy for the 21st century, arguing that reading and writing instruction must now encompass the ability to critically evaluate AI-generated text, recognize the hallmarks of synthetic content, prompt AI systems effectively and to understand the social and ethical contexts in which AI-generated language circulates.

12:30 p.m. 鈥� : This keynote by Adeel Khan, Founder & CEO of MagicSchool AI, makes the case that teacher expertise, relationships and professional judgment must guide technological change. Drawing on his experience building the popular platform, Khan will share unfiltered insights on what’s working and what’s not, offering a framework for evaluating AI tools through the lens of educator agency.  

2 p.m. 鈥� : This panel examines why so many school AI initiatives rely on tools that 鈥渏ust aren鈥檛 there yet.鈥� Panelists share case studies of implementations that stumbled, the lessons of those failures and the educator-driven, grassroots efforts that can move schools from dabbling with AI tools to using them for real instructional transformation. 

Thursday, March 12:

10 a.m. 鈥� : This featured panel convenes former Presidential Science Advisor Arati Prabhakar, Renaissance Philanthropy President Kumar Garg, Carnegie Learning VP of R&D Jamie Sterling and Bezos Family Foundation Chief of Staff Eden Xenakis to explore how bold learning goals can accelerate AI-driven innovation in education. They鈥檒l examine how 鈥渕oonshot-centered鈥� models can rally diverse innovators around a shared outcome and catalyze the funding needed to scale breakthroughs.

10 a.m. 鈥� : Dubbed the 鈥渢oolbelt generation,鈥� more than half of Gen Z respondents in a recent survey said they鈥檙e considering a skilled trade career. And schools are working to modernize career preparation, including by tapping immersive technology to expose students to in-demand skilled trades. This panel, moderated by The74鈥檚 Greg Toppo, will discuss how we can harness tech to engage students in learning while preparing them to successfully meet workforce demands.

11:30 a.m. 鈥� : This session offers a ground-level counternarrative to AI anxiety, presenting a community college and workforce development partnership in Cleveland that is using AI-powered tools and training to open new economic pathways for adults who were left behind by earlier rounds of technological change. Speakers will examine what equitable AI adoption looks like in a post-industrial city and what conditions made the initiative work.

11:30 a.m. 鈥� : Leaders from higher education, industry and workforce policy examine whether universities are structured to produce graduates who can thrive in a labor market being remade by AI. The panel will ask which degrees and credential pathways are producing AI-ready graduates, where institutions are falling behind, and what structural changes will move the needle most.

11:30 a.m. 鈥� : Directed by Scott Barnett, this feature-length documentary follows bestselling author James Patterson to the front lines of America’s reading crisis to examine how the Science of Reading 鈥� a vast body of evidence-based research 鈥� is changing how children are taught to read. A post-screening discussion with literacy researchers and classroom teachers will examine what the film gets right and what systemic change will actually require.

2 p.m. 鈥� : This workshop, conducted by two top officials with the Illinois-based Education Research and Development Institute, will offer practical AI tools that automate routine tasks, generate content, analyze data and simplify communication, freeing teachers to focus on students and strategy and reducing the risk of burnout.

2:30 p.m. 鈥� : This featured panel, with Martin McKay of Everway, Hello Sunshine CEO Maureen Polo and the Brookings Institution’s Rebecca Winthrop, draws on a landmark report spanning 50 countries to explore what it means to protect children’s cognitive, social and emotional development in an AI-saturated world. Speakers will move beyond the question of whether AI should be used in schools to ask how it can be designed to strengthen young people’s capacity to think, relate and thrive.