speaker-0: My advice would be lean into this AI moment with a lot of curiosity. Lean with pilots. Be slow to scale. Keep a high bar for scaling. No one has this all figured out. And the only way to figure it out is by trying it. You'll find use cases with your teachers and your community that others won't. And the moment you find those use cases, it unlocks that superpowers. But I you have to experiment with it. I I there's a great, there's a great team called Decoding Futures up in New York, and they they talk about learning AI is a little bit like learning a There's only so much you can learn just by reading about it. You have to practice it.
speaker-1: Today I'm joined by John Bailey, a non-resident senior fellow at the American Enterprise Institute. John has served as director of educational technology in both the Commonwealth of Pennsylvania and the U.S. Department of Education with a career that spans the White House, philanthropy, public policy, and technology. And that gives him a really rare view of what happens when ambitious ideas meet the reality of schools. So critical today. And so today we're going to be talking a little bit about what the last technology cycle can teach us about AI, the path for districts to turn new tools into stronger teaching and learning and ultimately stronger schools. So, John, welcome to Strongest School. Happy to
speaker-0: To have you here. my gosh, thanks for having me.
speaker-1: absolutely. So now you have worked on technology really at the intersection of technology and education for quite a while. I think kind of spanning kind of the internet years to to now generative AI. I love your thoughts on kind of how, you know, what's your mental model right now for thinking about how we're dealing with this this AI moment, this this cognitive revolution that we're really in?
speaker-0: Yeah, no, I've had, I mean, a remarkable privilege of being at the intersection of technology and education. I mean, going all the way back to the nineties and and seeing multiple waves of an effort to try to harness the immense capabilities of a lot of this technology to personalize education for students. And you know, and there I think, you know, it the the couple of lessons is we way overpromised and we underdelivered. And I I don't think our ambitions were wrong. It's just that I think we settled for too little and we just always fell a little bit short of being able to deliver that experience for students. And I I think in the process also frustrated a lot of teachers because I think a lot of teachers wanted these kind of tools. I think a lot of teachers wanted these opportunities for their students in their classrooms. And we just fell short a little bit. It I and I and I think there's that explains a little bit of some of the skepticism that we are facing right now, because I think a lot of teachers have heard this playbook before, but it feels very different. I I think the more you use some of these AI tools, you really do feel like there are capabilities here that are very different than the technologies I saw in the 90s and the 2000s, and even as recently as three years ago. And so I I think this a remarkable opportunity and challenge for us. To really lean into the moment and figure out how do we take the best of these AI tools and put it to work inside of our classrooms for teachers and for students.
speaker-1: Yeah, really interesting. It almost seems like d listening to that, that we've we've kind of shifted from kind of a capabilities deficit a little bit to kind of you mentioned kind of the the the the overpromised underdeliver to to maybe now like a diffusion or a skill set deficit. Is is that kind of how you might
speaker-0: I think it's true. I mean, if you think about it, like past technology waves, I mean, even going back to the printing press and books, like that, that helped to democratize access to to information. And the internet sort of connected all those sources of information. I think we always thought, like, if students had access to more information, better learning would happen. And that's true to an extent, but it it it's still not entirely true. And it turns out you need a lot of supports for teachers and a lot of supports for For students, so we leaned it a lot into personalized learning platforms. And I think again, the intent there and the ambition was right. I just think we fell short. And then all of a sudden, Gen AI comes onto the scene and it just feels immensely powerful. For the first time, you can ask any complicated question that you have, or or any information you're struggling with understanding, and get an infinitely patient. Tutor that'll just explain it to you in any way that you want. So if you want to learn about chain-weighted CPI in the style of someone who's obsessed with Taylor Swift or with football, it'll gladly give you a lot of examples using that. And so all of a sudden it takes these things that have been very inaccessible and hopefully makes it a little bit more easy to understand and scaffold your understanding around. And I think that's you know, just a new set of capabilities that we've never had before, something that can generate different examples in. Infinitely be patient with a teacher or infinitely be patient with a student opens up an enormous amount of opportunities, whether it's differentiated learning, whether it's helping a teacher with creating multiple activities for different students at different progressions inside their classrooms, or for creating coherence for just trying to help make sense of all the different pieces of data and information and curriculum and supplemental materials and other resources that a teacher has. All of a sudden they have a a really powerful AI tool that can help them make sense of that.
speaker-1: So it's so interesting as I as I hear you talk about this. We, you know, we have we're at a moment where the the possibilities of of AI to deliver, you know, kind of with you know, genius on the other end of a screen, right? On demand, put things into all kinds of different contexts, really provide all this access, right? So we're we're as We're as close as we've ever been to like kind of truly delivering on this like vision of personalized education, which everyone, you know, kinda can feels is absolutely right. And yet we're seeing this incredible kind of retention, concern. Yeah. Right. And and I don't want to put that in pejorative terms because there are real, real concerns about, you know, obviously workforce,
speaker-0: Four.
speaker-1: but also just is it effective? Is it a an effective way for a student to sit there and learn? and is that just a where we are on the on the adoption curve of this? But what what can we learn from maybe past or or just you know currently? Like what's your thoughts on that that kind of dichot dichotomy?
speaker-0: I mean AI, I mean these immensely powerful tools are coming out at a moment when, you know, there's a retrenchment in some ways from MedTech. And I I think some of the bans that we're seeing are the wrong answer to the right question. And I I I think there's rightfully Superintendents and parents and teachers too who are asking, you know, is this technology needed? And it it's coming in the wake of I think a lot of concerns around social media and cell phones and distractions that creates inside the classroom. I think it's also coming a couple decades into this ed tech experiment. And we've had a lot of experiments and and sadly a lot of them have failed, and that they failed students, they failed teachers, and and so I think there's a little bit of jadedness. And I understand that, but I I think. That's not an excuse for backing away from AI. I think it's keeping very high ambitions. I think it's keeping very high standards, and it's just making sure that we're using AI to answer very specific problems that are defined inside the classroom or inside of a school and and being clear on measuring the outcomes of that and not just measuring the adoption. I think this is the other thing we got wrong in past cycles was we looked at. know how many computers do we have inside of a classroom, how many classrooms were connected, how many students were using a particular type of tool. And we we ignored, I think to our detriment, whether any of those tools were impacting their learning and their teaching. And so I think we need to go into this AI era really clear that the only success criteria is that if it's improving teaching or learning Or whatever any other type of education outcome that that superintendent or principal or teacher is deciding, whether it's parent communication, whether it's helping with optimized school bus routes, whether it's helping with making sense of supplemental curriculum and their core curriculum so that coherence. But you have to be very clear on the outcome and the problem that you're trying to solve.
speaker-1: Yeah. I want to come back to a couple of those use cases in in just a minute, but th this is sounding a little bit like a a a term I think you coined, which is productive struggle. and so I'd love if you could unpack that a little bit for us, like what that concept means in the context of AI and then how we wanna kind of think think about that.
speaker-0: I productive struggle was not something I coined. I I it's it's the a term that's been around for a while and Bellwether education partners I think popularized it with a with the report last year. But the the notion of productive struggle is that it's learning is not just about getting to the answer, it's about the process of getting to the answer. And and part of what learning whether you're an adult or whether you're a student, is that struggle. It's the struggle with the first, second, third, and fourth draft. It's the editing and revising, getting feedback and incorporating it. It's working through the hard math problem. And so that productive struggle leads to durable education outcomes on the other side. You end up learning the material more, you can recall it more. And I think this is one of the greatest threats with AI is that all of a sudden you get an easy button, can give you the answer, can give you often a better answer. But you haven't learned in that process. And so I think this is one of the things that we're hearing a lot of concerns about from from districts is that yes, their students are doing better on assignments, but they're actually not learning. And it's because they're using these tools as a crutch, not as a means of actually sort of helping to be proactive in their learning.
speaker-1: That makes a ton of sense. So it's this idea of, you know, you know, you you write to think kind of kind of kind of concept, right? which which makes sense. Have have you seen evidence of how tools or how how districts are are are kind of grappling with this? How how do we, you know, how does a district leader think about the balance between they they want their students to be, particularly in the in the kind of the the upper grade levels, right? we'll we'll kind of leave out you know kind of lower elementary aside for a minute. But now they want students exposed to these tools. They know it's going to be important for their fruit future career, that they're incredible kind of again, that the this kind of cognitive access that you have. And yet how do they maintain that tension and create the the the the struggle that that they need.
speaker-0: No, it's a great question. I and I I think we're seeing lots of experiments across the country. one I just heard, you know, of a teacher that they they're not banning AI, but they they use kind of a red light, yellow light, green light for each of the activities. Or green light for the activity, students can use AI all they want. Yellow light, they have to use it, but disclose a little bit more or be a little bit more cautious in how they use it. Red light is like no AI can be allowed. And so with that. I think does is it empowers the teacher to sort of think about what's the right and appropriate way of using a portfolio or suite of AI tools for any kind of given assignment. And it gets away from sort of the b the blanket b bands. I I the another one that just came out in the last like week is David Deming up at Harvard has like an interesting approach. He calls it kind of like a barbell approach, which is like any assignment that he gives his students, initially he wants them not to use AI. He wants them to start thinking and working through it with just themselves. Then the second phase is really using AI to interrogate that, to research, to help you know, try out other types of frameworks, but use AI as a tool to help them really explore. kind of the issue. And then the third part is that they have to defend it. And this is the accountability part where you take away the AI and you sort of see whether or not the student can actually defend and articulate and and really explain what it was in terms of the project with it. And I kind of like that approach because there's a little bit of using AI as a tool, but really testing to see if it's improved the capabilities of the student as opposed to just they get to the right, the right answer. the other w reason I like this a little bit is that in In the the Frontier Labs, they will often talk about this thing called chain of thought. And chain of thought is when you give a really complicated project or or question to an AI system, it shows its chain of thought. It shows how it reasoned through to get to the answer that it gives you on the other side when it produces it. And so, in some ways, this is actually bringing chain of thought reasoning into the classroom. It's asking students to show and be transparent about every single step along the way about where were they thinking, how did they use the tools. What did the tools produce that was accurate or not accurate, then ask the student to defend it? So there's a kind of a remarkable I think, opportunity here to sort of adopt chain of thought reasoning and bring it inside the classroom and help it to hopefully create the productive struggle that we want our
speaker-1: And I love those examples and and it does kind of bring up this idea of like how how assessment will really change you know in the age of I AI, right? How how is that, you know, or teachers the the the idea of just a a test anymore is is gonna be challenged, right? So the idea of using discourse, you know, or back to blue books, you know, whatever, whatever it might be.
speaker-0: Yeah, I think I think it's all all the above. I think we'll we'll go back to blue books. I think we'll go back to student presentations without tech. I also think you all of a sudden have this opportunity for tech to make sense of a lot of student works that we never had that chance of before. You think about, you know, student notes and student papers, it all sort of sits inside of a drive somewhere and And in the past, it would have been too much to ask a teacher to review all that and in a coherent way try to try to assess it. And now all of a sudden you could have AI agents that are just constantly sort of assessing student work. I'm really interested in I think some of the new multimodal capabilities of AI that can not just read what a student wrote, but also gauge how a student presents and speaks. It gives them feedback in that way. I think that's gonna be remarkable for teachers because now teachers can get quality based quality, the teacher quality can be measured and given feedback and just in time coaching just based on AI analyzing video and audio. So I it's been one of my favorite uses of Chat GPT lately, is I'll f I'll fire up the the voice mode and I'll practice speech, I'll practice remarks and have it coach me in real time. And it's been it's been powerful to just have that in my pocket anywhere that I go that I can get that just in time coaching. So and it's a bit of assessment. It gives me good feedback. But I I I think we're in gonna be in a new era where there's all sorts of new assessment mechanisms that this AI tools will be able to give us.
speaker-1: Yeah, it's so interesting. I I mean I think one one word that comes to mind is intentionality, right? That that districts at the at the appropriate grade levels really need to be thinking about how to be intentional in the way that they're educating students about AI and the way that students are are interacting with the AI. But but we shouldn't I I mean I I think maybe a larger point is that we shouldn't kind of throw the baby out with the bathwater, right? We we the the the concern is real, but I think there's all kinds of, you know, and I I've been hearing myself all kinds of very interesting Use cases of increasing accessibility through AI for people that learn differently. The multimodal thing right now, maybe a student is an auditory learner, a visual learner. And the ability to, you know, as long as they're not kind of just getting the easy button to get the answer, the ability to have kind of a dialogue and and to seize these issues from different different directions can be absolutely incredible.
speaker-0: It can be, yeah. And I mean there's all sorts of ways now to engage with the material and to be and to be quizzed on it. aga again, you could just ask the voice mode and any of these tools to to just do a just in time quiz and it'll say something and it'll quiz you on it and it'll it'll give you coaching based on not just did you get it right, but how did you explain it? Did you stutter? Did you use ums and hus and again that's like a pretty powerful capability right in your pocket. I also you said something really important too. You said we shouldn't throw the baby out with the bathwater and I I I think that is another temptation right now is to think that AI changes everything. And it means everything inside of education needs to change too. We need new AI literacy and new prompting classes. I think that some of that is true on the margins, but in some ways it makes the fundamentals even more important. And you said before, and I I've said this too, that writing is thinking. And and that is particularly true as it relates to using these AI tools, that the the capabilities only get unlocked based on the sophistication of the prompt that you give it. And the and prompts aren't magical, they're they're English language written in a very logical way, in a coherent way, in an articulate way. And so, in many respects, it makes all the work we've been doing on ELA. even more important because it's the students who are proficient in ELA and in English and writing coherently and logically and concisely, they're the ones that are going to be able to unlock the prompts and the capabilities of these AI tools. And so in a weird way, the programming layer of this new era of AI isn't just math and isn't computer programming. It's also ELA. And so we should be leaning more into that now to help make sure that the this generation of students are prepared for the workforce they'll be entering.
speaker-1: Yeah. I love I I've always been struck and and really liked the concept that Ethan Mollock kind of came out with of co-intelligence, right? as kind of thing about the the way of thinking about how we're gonna interact and and and but it is you know it seems to me that we're this moment where, you know, kind of the technology and the pedagogy are are kind of, you know, they need to keep pace with each other. And then this goes right back to kind of a we started, like kind of a diffusion. story, right? How what's the r you know, it kind of feels like we're that's what we're grappling with most is how do we get the this incredible technology appropriately into the into the hands of educators and and ultimately into students.
speaker-0: No, I think that's right. It's like thinking appropriate. And also, you know, back we were talking before about different waves of technology. When the internet first came out, every district created an acceptable use policy. And this was this was very much about empowering local schools with their local communities to figure out what was acceptable, what was the right way of using the internet, using resources found on the internet, of using Wikipedia. There was a lot of concerns similar back then too, that students were just cut what they cut and paste what they found in Wikipedia into their papers. And so acceptable use policies allow communities to really kind of pace themselves with the technology in a way that's like values aligned and also aligned with their their parents and with their students. I think we need something similar to that right now. it's also you know, it's it's thinking about there's different types of AI. Like you know, there's these relational chatbots call it character AI, has no place inside the classroom. And I think you know, those. Those you can be pretty comfortable with banning. I think you have some of these general AI tools, ChatGPT, Gemini, you have you know, Claude, obviously. And they're, you know, you may not want to just completely give every single student access to that. Maybe you want to give your teachers access to that because it could be a powerful teaching tool. Maybe you want to hold that back from students or only allow students to use it in certain situations. And then you have purpose-built AI tools, and that is everything from PowerSchool to You know, brisk teaching, there's just a lot of these new AI tools that are coming out that have education safeguards built into it that have been designed with teachers and students in mind. And there, I think that that could be you can lean into that with a little bit more permissiveness. But also I think you have to be very clear on the problem you're trying to solve and and what does quality look like? How do we know something is improved by using that tool? And that is gonna be a a decision that really comes down to principals and to teachers to be. very clear about that and make sure that they're measuring the outputs and the outcomes.
speaker-1: And that brings our back to data, right? Like w ha you know, we need the data to actually tell us you know, ha what's what's the impact of these of these tools. interestingly, total aside, that we just did a a survey of over 350 district and school leaders for kind of the pack to school season, specifically asking like how satisfied they are with their ability to measure the impact technology.
speaker-0: yeah.
speaker-1: and and the news isn't great, right? Like very low satisfaction level. Yeah. really a lot of just using anecdotal feedback, very, very low in terms of actually being able to use kind of, you know, true efficacy feedback loops. And I think that goes down to like, you know, the A, there's a, you know, cause and effect is really hard to prove when there's a lot of of other factors, but also just the data that they might have available to them. So definitely some some work to do in terms of how do we give district and school leaders like better tools to understand the impact of these tools and how can we kind of accelerate the feedback loops on some of those things.
speaker-0: No, I agree. There's just one point on that, because I think it's it's it's an important point, it's a powerful point. But we've spent decades trying to give educators and principals and superintendents more data. We've done it through dashboards and we've done it with assessment data. And and I mean, in some respects, school systems and classrooms are swimming in data. And it the hard part has been making sense of it and seeing pattern recognition, seeing the coherence kind of within it and doing something with it. And again, maybe that is the killer app for AI right now. Maybe it has none. Nothing to do student facing. Maybe it is just helping teachers and superintendents and principals make sense of this data that they're they have already. And in if that is all we do, that could be immensely, immensely powerful. But but again, I think it's like it's trying to think through the tools and think through what is it that we're trying to improve and what is the best ways of collecting the data or analyzing the data to see if the patterns are there to show that we're improving.
speaker-1: Well y you make a great point. And it's a great it's a great kind of pivot point for us, right? Like so much of that energy in the focus and and understandably so for kind of AI in K twelve has been kind of the student facing the the the personalized tutor. but we're not thinking maybe as much. generally speaking on the impact for administrators, for teachers at a as a productivity tool, as an insights tool, and maybe ultimately as a or or eventually as an automation tool when we start thinking about agentic. I'd love you, we you met you kind of passed through very quickly a few use cases earlier in our conversation. Love if you come back to those, maybe talk about a couple where you're seeing some some interesting or the the the the highest potential right now.
speaker-0: I have struggled for last three years to try to figure out like what is the right metaphor for explaining what AI is. Is it a tool? Is it a service? And then I I I was with a chief technology officer out in Silicon Valley, and he had this great, he talked about AI giving people superpowers. And I was like, you know what? That actually is pretty accurate and pretty clever. And and I want every single teacher to have these superpowers because already we're asking great teachers to do almost the near impossible. They're we're supposed to go into a class of 25, try to understand where all those students Students are, where they're struggling, try to develop some sort of coherence amongst all the different applications and supplemental and core curriculum and all these other materials they have. They're trying to differentiate instruction. They're trying to make a sense of why some students might be chronically absent. We're asking them to do a Herculean job. And all of a sudden, now we might be able to give them some AI superpowers to help make some of that work easier, to help free up some of their time, whether it's on administrative tasks. Or to give them deeper insights. And and so I what I want is for every teacher to have those superpowers. However, they need to deploy them inside the classroom. Some of it might be developing instructional materials. We have a lot of teachers that are teaching out of field. They're amazing teachers, but they just don't have necessarily the depth and content knowledge. And so maybe what they need is a agentic teaching assistant that helps them get up to speed on some of the content before they go into the classroom. Maybe it's new types of assessment. Gentec AI is helping to just give a sense of where all the students are as they're coming into the classroom that day. So I just I think this idea of getting giving teachers superpowers. And once teachers have superpowers like that, they never want to give it up. And and I think that is the way to sort of start, I think with pilots. I think that's the way to start building some of the early AI use cases. But also I think this is perfect too on the administrative side. Like again, not even doing anything student-facing all of a sudden now. Principals and superintendents have an immense powerful AI tools that can help them with reaching parents, with better parent communication, with language translation, with helping to develop coherence amongst their schedules. There's a whole bunch of different use cases there just to help make school districts operate and function better for their students, for their teachers, and for their parents.
speaker-1: You have thousands of administrators and teachers vigorously shaking their heads up and down right now, nod nodding in agreement, particularly on the workload challenge. And and that's a that's a big focus for us for sure, is thinking about great, how do we, you know, what are some applications where we can kind of bring and you you hit on a couple which are great. What what are your, you know, how do you think about kind of the underlying kind of the the infrastructure requirements or are some of the ways that we can kind of you know. A AI, of course, is so dependent on kind of context and data and grounding. and so I love your thoughts on some of the ways that that, you know, districts can kind of think about setting themselves up for success. The way that maybe a lot of maybe enterprises that are already on the AI wave have been thinking the last couple of years and maybe have the flexibility to move, you know, at at whatever pace they're able to move to kind of put some of the grounding in place to to become more AI ready, more agentic, if you will, and and how how do you see that playing out in the district?
speaker-0: No, I think it's an important point you raised because these new models, it's been pretty much the case since ChatGPT four, but I mean, especially with these new models, the way you unlock so much of the capabilities is by giving them a lot of context. And it's not just giving them a better prompt or better direction, even though that's important. Prompts again are about defining what it is you want the AI to do, giving it guidelines about what good and what bad doesn't look like what good looks like and also what bad. bad outcomes would look like. And then a lot of context. And what I have found is that these these models are insatiable for context. And so if you think about it, either the context of your classroom as a teacher, you know, you have context with your school district, with, with school district policies, with school district curriculum. You have some flexibility. You've chosen a bunch of tools yourself. You're using different types of supplemental resources. you have a different set of lessons that you've done in the past and you have lessons that you have planning come out. All that is context. And the more context you give it, usually the better the output is on the other side of it. And I I think school districts and and schools can also lean into this too by helping it create skills and and other kinds of data sets that help the teachers AI agents and the teachers AI prompts with you know sort of pre-loaded set of context that they can just sort of plug and play inside of it. But I think context is going to be immensely important going forward and articulating that and getting that in a pretty rigorous way is going to help us make sure there's higher quality outputs on the other side.
speaker-1: Yeah. Yeah. And we were talking earlier, you were kind of mentioning this idea that we're in this you know, kind of inflection moment of going from AI as as a kind of a thing that you talk with. We're we're all kind of familiar with the chat GPT kind of chatbot experience, to a thing that that is taking action, that is doing things, more of the agentic model. Maybe you could talk a little bit about that and kind of give you give your view on that for for kind of the folks in our audience.
speaker-0: Yeah, I think it's it's for folks listening to this, it's it's you have to think of AI as like in sort of two different categories. There's the AI that I think most people have been using right now, and that's ChatGPT and Gemini and Claude, and it's it's almost like a super Google. Like they're using it to give to get an answer. Maybe you go back and forth with it a couple different times. And the task is like writing something for the most part, maybe generating an image, maybe generating a spreadsheet, but that's pretty much it. The era that we're moving into, and it is so important to understand because of how seismic of a shift is, is agents are very different. Agents are AI systems and models that act somewhat autonomously. So you give it a goal, and then it reasons on the best way of executing that goal. And it can work over longer horizons, and it can use more tools, and it can produce more things. And so Some of the AI agents that I've used have worked for 30 hours at a time on a problem. it'll build things, it'll build spreadsheets, it'll build just-in-time applications that you can use you know, for your work. Or I can imagine in the future, you know, teachers are gonna task an AI agent something and it'll develop. new resources for them to use inside the classroom as well. But agents are very different because you're prompting it, but you almost are acting like a manager. A agents are showing up in the workforce, in the healthcare sector, almost as employees. And so what you are doing is you're managing a team of AI agents and you're managing a team of humans. And that means you have to be very clear in terms of the task that you're assigning it. You have to be very good at reviewing the output of it and giving feedback and saying we're at mid met the mark and where it missed the mark. but you're managing and it's a very different type of skill set. but to give you an example, OpenAI just announced last week that they had a group of 10,000 agents solve a mathematical challenge that had that humans have not been able to solve for a hundred years. And what you have to understand is that OpenAI, yes, has access to enormous amounts of compute and sophisticated models. They had 10,000 agents working together to solve this problem. I think in two years, you're gonna have teachers that have the capabilities of tasking hundreds, if not thousands, of agents to work on a problem inside their classroom. That could be helping them with developing new resources. It could be helping them with figuring out what is the best way to help John understand this calculus problem since he seems really stuck on it. But I I think it's gonna be an amazing amount of capability boosters for teachers, but it's also very different because now you're not just prompting, you're also managing. And a lot of the ways these Prompting and giving a task to an agent looks is almost like a job description. Or it's almost like for a school superintendent, they could appreciate this. It's a scope of work. It's it's outlining what you want the agent to do, what success looks like, how you'll evaluate its quality, and then giving the agent the time and space and context to to actually work on the problem and come back to you to read.
speaker-1: view it. Yeah, that's amazing. And I can just already think of you know a half dozen you know, really important domains within a district where this kind of agentic approach could just kind of really alleviate so much kind of administrative burden on superintendents and on the cabinet, whether it's thinking about, you know, early warning indications for attendance, whether it's identifying, you know, an intervention earlier for a particular student or group of students. So really, really incredible to kind of think about where that where that goes. And and it has nothing to do with, you know, necessarily like AI, student facing AI in the classroom, it's just kind of a really thinking about how a district really transforms the way it operates. you know, use use
speaker-0: No, I think that's right. It's also thinking, you know, you almost you have there there's a lot of risks of anthemop anthromorpizing some of this too. But like, but but you almost want to think about like who is the team you want working on a problem. And so like I I think right now a lot of people are like, how do I get one of these models to be a really good tutor? And I think that where this is evolving is that you're gonna have a group of agents working collectively as a tutor. So you're gonna have a productive struggle agent. Whose whole job is to make sure that the student is actually engaged in a productive struggle. You're gonna have a coherence agent that's gonna make sure whatever the output is is all aligned with the district curriculum, supplemental curriculum, with the standards of the state. And you're gonna have a student engagement agent that really knows the student really well and is gonna say, okay, the best way of teaching John is by using football examples. And they're gonna work together as a team to help deliver. the experience for that student or for that teach that teacher. But I think it's a very different way of sort of thinking about things. I know it sounds like science fiction right now, but I'm seeing examples of this in healthcare. Definitely seeing examples of this showing up in the corporate enterprise. And so I think probably in the next year you're gonna start seeing agents showing up much more inside the classroom. And it's gonna give teachers a new set of superpowers as part of their portfolio.
speaker-1: So how do we think? Let's go back to a little bit where we started. We have, you know, and and we totally understand where where these are coming from. We have major districts that are banning AI tools at various grade levels. And there is a real risk that, you know, in the the again, the very real concern that district and and families may have about kind of that the the AI taking over, that the experiment implementation that's going to lead us to to those approaches, you know, might might be missed. I guess like how do we help, where's the balance? How do we help kind of you know provide the leadership for for districts to kind of see their way through this.
speaker-0: I think well, districts need to look at this as they can only move at the permission of trust. And the most important trust is gonna be with their parents. And I I think it's recognizing that parents are very anxious right now. They're anxious about again what's just happened with social media and with cell phones. They're anxious, I think, coming out of COVID, where they saw a lot of tech and not a lot of learning. And they saw a lot of harm. They saw a lot of their students. And we see it in the national education data with the PISA data. We we have just seen a generation of students that are behind. And and And parents rightly or wrongly are blaming a lot of that on the tech experience that happened. And so I think it just
speaker-1: Here comes AI, right.
speaker-0: like Lucy and the football from the peanut in the the comic strips where, you know, every single time they go to Charlie goes to kick the football, Lucy pulls the football away. And I think a lot of teachers, a lot of parents have said, we've heard this before. And so I I I I I think it's meeting them where they are and acknowledging that. I think it's setting up pilots to build that trust. I think it's also realizing that again, parents are hearing a lot of concerns about AI in the economy. They're hearing about AI disrupting jobs. And so, you know, they're trying to make sense of this in society, much less like deploying this inside the classroom. So I I would start first with trying to build trust with parents. I would start with very small focused pilots. And again, you have to be very specific and explicit with the problem you're trying to solve and what does good look like on the other side of it. and then also I I think there's a lot of room for districts to experiment with again, nothing that's student facing, but instead with just looking at some of their administrative and operations thing and how do they improve just the functioning aspect of of school systems. But I would sort of start there and and build trust as as they go. And just and also lastly, I would say be very explicit about what AI is not gonna do. Like I I think handing over decisions about grading. handing decisions over as whether or not a student su should progress to another class or to the next grade level over to AI. Those feel very high risk. And and so I would always make sure humans are in the loop, teachers are in the loop as part of making those decisions, and where AI is a tool and a supplement to it, not actually driving the decision.
speaker-1: Yeah. Moving at the speed of trust. That's a great that's a great line. that that it's the only way, right? Yeah, that we can do it. And and and I think, you know, honestly and and and I and I'm confident like the the promise is too great, like the intelligence is too amazing that we will eventually get there and and and some of the the vision that you're painting in terms of the this kind of team and and how it's gonna be applied to to really help a student be at the the highest point of their kind of contribution. I know we'll we'll get there in the end, but I I'm sure it'll be in some fits and starts. But if we do nothing else other than alleviate some of the burden on administrators, on teachers, give them capacity kind of back in their day to, you know, in the case of administrators, spend more time with a teacher, in the case of a teacher spend more time with the students. I mean, that will be in and of itself just a huge, a huge, huge win. Yeah. well, amazing. Love, you know, kinda s take a step back. Like what and and you might have already answered a little bit, but maybe maybe go further, like you know, five years from now, we we see this vision. What will what will district leaders have have done right to to get us there? Right. What what are the what are the steps?
speaker-0: I think if we have done this right, I think district leaders will have started with the first thing we need to do is build trust with teachers and with parents. because trust, you can't do anything. I think the other thing we'll have gotten right this time that we missed in previous waves is that we were explicit in the goals. We were explicit in what we were trying to improve. and we worked backwards with what are the best ways of using AI to test to see if we can improve the outcomes or capabilities on those. But if we're not ruthlessly focused on the goal and the problem that we're trying to solve, I what I worry about in some five years, we'll have the opposite. We'll have we have a lot of AI tools in the classroom. We'll be judging how many students are using it, how many lesson plans it's generated. And we'll have missed that it doesn't really matter. What matters is whether or not student learning increased, whether student engagement improved, and whether teachers had more hours that they could reinvest back in relationships and with giving feedback in their students. And so, but I think we have to keep in a laser-like focus on improving outcomes and having really clear measurable goals and problems that we're trying to solve.
speaker-1: Yeah, that makes a ton of sense. Let me shift gears for a minute. We've talked a lot about kind of application of of AI in in schools and different different use cases, but let's talk about kind of the the broader impact kind of from a workforce perspective. And I don't know if you have some thoughts on you know, current moment as it relates to how we provide how we generate a kind of a an AI ready workforce, right? Like it's it's, you know, we're feeling it in in our professional lives every day, but it's kind of changing the nature of work. And so curious if you have some thoughts on on how, you know, districts should be thinking about preparing students, how the system should be thinking about you know, those those types of outcomes for for students.
speaker-0: yeah, it's a great question. I I and first of all, there is such a rigorous debate right now about AI's impact and its disruptive impact on the workforce. And I think everyone's trying to make sense of it. It is hard because the models keep changing so rapidly, and trying to gauge whether or not like a ChatGPT five or you know. Gemini 2.5 has an impact on certain tasks inside the workforce. It's very different than as the agents are showing up and are capable of working longer and doing a greater range of different tasks. So everyone is trying to make sense of this. I I think the future is definitely ingenic. And what that means is it makes K-12 education even more important. And it and for the basics that we were just talking about, that it is impossible to manage a team of agents. if you can't clearly articulate what you want those agents to do. And so ELA is become the new programming layer and is more important now than it probably ever has been in the past. It makes math very important, not from a coding standpoint, because math teaches you how to think and how to reason and how to work through logic. And often the way you manage a team of agents is thinking logically and evaluating the output that they give you. And so math and reading become more important, not less important in this world. it also means every new entry-level job that's created, I think, in the future becomes higher skilled. And it's just because Capabilities of these AI tools, these AI models, and agentic AI means it can keep doing lower and lower skilled. It's going to eventually become higher skilled, which means every new job is gonna become higher skilled and needs a more skilled workforce for it. So K-12 education is even more important now than it has been in the past. but we have to be making sure that they're graduating proficient in reading, writing. in a proficient in math because those are gonna be the core skills to unlock this agentic future.
speaker-1: Yeah. It's so well said and and it kind of goes back to this idea like this this renewed focus on on learning how to learn and learning how to think as being kind of you know the critical skills in addition to you know all of the skill trades that we're gonna now need that out of kind of we've got resurged in in in in importance in our in our economy.
speaker-0: Trades have always been important. They're even more important now. you know, I also I I suspect there there's a economic theory called Jevron's paradox, which is when you make something cheaper, demand for it doesn't go away, demand goes up. And I think we're seeing that in a number of different fields right now where yes, AI is eliminating some computer programming professions, but we're actually seeing a rise in the increase in the number of computer engineers and computer programmers. And it's because all of a sudden non-coders like myself are writing programs and writing solutions to problems I have. But at some point you hit a wall. And who do you call when you you hit that wall? You need an engineer, but that engineer has to be higher skilled. Yeah. And so in many respects, I think we're gonna lower the floor because of AI ha giving individuals, teachers, students, and the average American worker access to this expertise, they could do more work they could never do before because it's now all of a sudden cost effective and they have a a team of agents that are available to them to do it. But that's gonna drive demand, I think, for a lot of these other professions. And so Yeah, I I'm a little bit more optimistic on this. I I I think there's a lot of folks who are concerned about AI is gonna massively disrupt the workforce. I think there's another scenario here where actually we have labor shortages. And and again, all that comes back to education. And we have to make sure students are leaving high school with the fundamentals to prepare for any of any of those scenarios in the world.
speaker-1: Yeah, absolutely. Well, last question. you have the opportunity to give a K twelve district leader, a superintendent, advice on how they can build a stronger school today. what would what what's your advice?
speaker-0: I would my advice would be lean into this AI moment with a lot of curiosity, lean with pilots, be slow to scale, keep a high bar for scaling, but lead with pilots. I no one is f no one has this all figured out. And the only way to figure it out is by trying it. You'll find use cases with your teachers and your community. that others won't. And the moment you find those use cases, it unlocks that superpowers. But I you have to experiment with it. I I there's a great, there's a great team called Decoding Futures up in New York, and they they talk about learning AI is a little bit like learning a foreign language. There's only so much you can learn just by reading about it. You have to practice it. And so, what I would encourage is superintendents and principals to really start practicing with it. Just try it. If you had, what would you give a remote worker as a task? Just try that. And and whatever that email is that you would write to that remote worker, use that as your prompt. That begins unlocking use cases. and then the moment you start unlocking one, you start seeing a hundred. But I I think this is a remarkable moment for superintendents and for principals because they get this immense capabilities. that are coming in the form of AI and agentic AI in the future. And it's gonna help them unlock a whole nother generation of teaching and learning that I I I think can lead us into an amazing set of prosperity. But again, it's gonna be up to the leaders to harness it and to use it in a strategic way to to make that happen.
speaker-1: Well, on that note of abundance and positivity. John, thank you very much for joining Stronger School. Really appreciate you joining.
speaker-0: Thanks for having me. Thanks for running.
speaker-1: Yeah, you bet. All right. Thanks everyone.
speaker-1: Today I'm joined by John Bailey, a non-resident senior fellow at the American Enterprise Institute. John has served as director of educational technology in both the Commonwealth of Pennsylvania and the U.S. Department of Education with a career that spans the White House, philanthropy, public policy, and technology. And that gives him a really rare view of what happens when ambitious ideas meet the reality of schools. So critical today. And so today we're going to be talking a little bit about what the last technology cycle can teach us about AI, the path for districts to turn new tools into stronger teaching and learning and ultimately stronger schools. So, John, welcome to Strongest School. Happy to
speaker-0: To have you here. my gosh, thanks for having me.
speaker-1: absolutely. So now you have worked on technology really at the intersection of technology and education for quite a while. I think kind of spanning kind of the internet years to to now generative AI. I love your thoughts on kind of how, you know, what's your mental model right now for thinking about how we're dealing with this this AI moment, this this cognitive revolution that we're really in?
speaker-0: Yeah, no, I've had, I mean, a remarkable privilege of being at the intersection of technology and education. I mean, going all the way back to the nineties and and seeing multiple waves of an effort to try to harness the immense capabilities of a lot of this technology to personalize education for students. And you know, and there I think, you know, it the the couple of lessons is we way overpromised and we underdelivered. And I I don't think our ambitions were wrong. It's just that I think we settled for too little and we just always fell a little bit short of being able to deliver that experience for students. And I I think in the process also frustrated a lot of teachers because I think a lot of teachers wanted these kind of tools. I think a lot of teachers wanted these opportunities for their students in their classrooms. And we just fell short a little bit. It I and I and I think there's that explains a little bit of some of the skepticism that we are facing right now, because I think a lot of teachers have heard this playbook before, but it feels very different. I I think the more you use some of these AI tools, you really do feel like there are capabilities here that are very different than the technologies I saw in the 90s and the 2000s, and even as recently as three years ago. And so I I think this a remarkable opportunity and challenge for us. To really lean into the moment and figure out how do we take the best of these AI tools and put it to work inside of our classrooms for teachers and for students.
speaker-1: Yeah, really interesting. It almost seems like d listening to that, that we've we've kind of shifted from kind of a capabilities deficit a little bit to kind of you mentioned kind of the the the the overpromised underdeliver to to maybe now like a diffusion or a skill set deficit. Is is that kind of how you might
speaker-0: I think it's true. I mean, if you think about it, like past technology waves, I mean, even going back to the printing press and books, like that, that helped to democratize access to to information. And the internet sort of connected all those sources of information. I think we always thought, like, if students had access to more information, better learning would happen. And that's true to an extent, but it it it's still not entirely true. And it turns out you need a lot of supports for teachers and a lot of supports for For students, so we leaned it a lot into personalized learning platforms. And I think again, the intent there and the ambition was right. I just think we fell short. And then all of a sudden, Gen AI comes onto the scene and it just feels immensely powerful. For the first time, you can ask any complicated question that you have, or or any information you're struggling with understanding, and get an infinitely patient. Tutor that'll just explain it to you in any way that you want. So if you want to learn about chain-weighted CPI in the style of someone who's obsessed with Taylor Swift or with football, it'll gladly give you a lot of examples using that. And so all of a sudden it takes these things that have been very inaccessible and hopefully makes it a little bit more easy to understand and scaffold your understanding around. And I think that's you know, just a new set of capabilities that we've never had before, something that can generate different examples in. Infinitely be patient with a teacher or infinitely be patient with a student opens up an enormous amount of opportunities, whether it's differentiated learning, whether it's helping a teacher with creating multiple activities for different students at different progressions inside their classrooms, or for creating coherence for just trying to help make sense of all the different pieces of data and information and curriculum and supplemental materials and other resources that a teacher has. All of a sudden they have a a really powerful AI tool that can help them make sense of that.
speaker-1: So it's so interesting as I as I hear you talk about this. We, you know, we have we're at a moment where the the possibilities of of AI to deliver, you know, kind of with you know, genius on the other end of a screen, right? On demand, put things into all kinds of different contexts, really provide all this access, right? So we're we're as We're as close as we've ever been to like kind of truly delivering on this like vision of personalized education, which everyone, you know, kinda can feels is absolutely right. And yet we're seeing this incredible kind of retention, concern. Yeah. Right. And and I don't want to put that in pejorative terms because there are real, real concerns about, you know, obviously workforce,
speaker-0: Four.
speaker-1: but also just is it effective? Is it a an effective way for a student to sit there and learn? and is that just a where we are on the on the adoption curve of this? But what what can we learn from maybe past or or just you know currently? Like what's your thoughts on that that kind of dichot dichotomy?
speaker-0: I mean AI, I mean these immensely powerful tools are coming out at a moment when, you know, there's a retrenchment in some ways from MedTech. And I I think some of the bans that we're seeing are the wrong answer to the right question. And I I I think there's rightfully Superintendents and parents and teachers too who are asking, you know, is this technology needed? And it it's coming in the wake of I think a lot of concerns around social media and cell phones and distractions that creates inside the classroom. I think it's also coming a couple decades into this ed tech experiment. And we've had a lot of experiments and and sadly a lot of them have failed, and that they failed students, they failed teachers, and and so I think there's a little bit of jadedness. And I understand that, but I I think. That's not an excuse for backing away from AI. I think it's keeping very high ambitions. I think it's keeping very high standards, and it's just making sure that we're using AI to answer very specific problems that are defined inside the classroom or inside of a school and and being clear on measuring the outcomes of that and not just measuring the adoption. I think this is the other thing we got wrong in past cycles was we looked at. know how many computers do we have inside of a classroom, how many classrooms were connected, how many students were using a particular type of tool. And we we ignored, I think to our detriment, whether any of those tools were impacting their learning and their teaching. And so I think we need to go into this AI era really clear that the only success criteria is that if it's improving teaching or learning Or whatever any other type of education outcome that that superintendent or principal or teacher is deciding, whether it's parent communication, whether it's helping with optimized school bus routes, whether it's helping with making sense of supplemental curriculum and their core curriculum so that coherence. But you have to be very clear on the outcome and the problem that you're trying to solve.
speaker-1: Yeah. I want to come back to a couple of those use cases in in just a minute, but th this is sounding a little bit like a a a term I think you coined, which is productive struggle. and so I'd love if you could unpack that a little bit for us, like what that concept means in the context of AI and then how we wanna kind of think think about that.
speaker-0: I productive struggle was not something I coined. I I it's it's the a term that's been around for a while and Bellwether education partners I think popularized it with a with the report last year. But the the notion of productive struggle is that it's learning is not just about getting to the answer, it's about the process of getting to the answer. And and part of what learning whether you're an adult or whether you're a student, is that struggle. It's the struggle with the first, second, third, and fourth draft. It's the editing and revising, getting feedback and incorporating it. It's working through the hard math problem. And so that productive struggle leads to durable education outcomes on the other side. You end up learning the material more, you can recall it more. And I think this is one of the greatest threats with AI is that all of a sudden you get an easy button, can give you the answer, can give you often a better answer. But you haven't learned in that process. And so I think this is one of the things that we're hearing a lot of concerns about from from districts is that yes, their students are doing better on assignments, but they're actually not learning. And it's because they're using these tools as a crutch, not as a means of actually sort of helping to be proactive in their learning.
speaker-1: That makes a ton of sense. So it's this idea of, you know, you know, you you write to think kind of kind of kind of concept, right? which which makes sense. Have have you seen evidence of how tools or how how districts are are are kind of grappling with this? How how do we, you know, how does a district leader think about the balance between they they want their students to be, particularly in the in the kind of the the upper grade levels, right? we'll we'll kind of leave out you know kind of lower elementary aside for a minute. But now they want students exposed to these tools. They know it's going to be important for their fruit future career, that they're incredible kind of again, that the this kind of cognitive access that you have. And yet how do they maintain that tension and create the the the the struggle that that they need.
speaker-0: No, it's a great question. I and I I think we're seeing lots of experiments across the country. one I just heard, you know, of a teacher that they they're not banning AI, but they they use kind of a red light, yellow light, green light for each of the activities. Or green light for the activity, students can use AI all they want. Yellow light, they have to use it, but disclose a little bit more or be a little bit more cautious in how they use it. Red light is like no AI can be allowed. And so with that. I think does is it empowers the teacher to sort of think about what's the right and appropriate way of using a portfolio or suite of AI tools for any kind of given assignment. And it gets away from sort of the b the blanket b bands. I I the another one that just came out in the last like week is David Deming up at Harvard has like an interesting approach. He calls it kind of like a barbell approach, which is like any assignment that he gives his students, initially he wants them not to use AI. He wants them to start thinking and working through it with just themselves. Then the second phase is really using AI to interrogate that, to research, to help you know, try out other types of frameworks, but use AI as a tool to help them really explore. kind of the issue. And then the third part is that they have to defend it. And this is the accountability part where you take away the AI and you sort of see whether or not the student can actually defend and articulate and and really explain what it was in terms of the project with it. And I kind of like that approach because there's a little bit of using AI as a tool, but really testing to see if it's improved the capabilities of the student as opposed to just they get to the right, the right answer. the other w reason I like this a little bit is that in In the the Frontier Labs, they will often talk about this thing called chain of thought. And chain of thought is when you give a really complicated project or or question to an AI system, it shows its chain of thought. It shows how it reasoned through to get to the answer that it gives you on the other side when it produces it. And so, in some ways, this is actually bringing chain of thought reasoning into the classroom. It's asking students to show and be transparent about every single step along the way about where were they thinking, how did they use the tools. What did the tools produce that was accurate or not accurate, then ask the student to defend it? So there's a kind of a remarkable I think, opportunity here to sort of adopt chain of thought reasoning and bring it inside the classroom and help it to hopefully create the productive struggle that we want our
speaker-1: And I love those examples and and it does kind of bring up this idea of like how how assessment will really change you know in the age of I AI, right? How how is that, you know, or teachers the the the idea of just a a test anymore is is gonna be challenged, right? So the idea of using discourse, you know, or back to blue books, you know, whatever, whatever it might be.
speaker-0: Yeah, I think I think it's all all the above. I think we'll we'll go back to blue books. I think we'll go back to student presentations without tech. I also think you all of a sudden have this opportunity for tech to make sense of a lot of student works that we never had that chance of before. You think about, you know, student notes and student papers, it all sort of sits inside of a drive somewhere and And in the past, it would have been too much to ask a teacher to review all that and in a coherent way try to try to assess it. And now all of a sudden you could have AI agents that are just constantly sort of assessing student work. I'm really interested in I think some of the new multimodal capabilities of AI that can not just read what a student wrote, but also gauge how a student presents and speaks. It gives them feedback in that way. I think that's gonna be remarkable for teachers because now teachers can get quality based quality, the teacher quality can be measured and given feedback and just in time coaching just based on AI analyzing video and audio. So I it's been one of my favorite uses of Chat GPT lately, is I'll f I'll fire up the the voice mode and I'll practice speech, I'll practice remarks and have it coach me in real time. And it's been it's been powerful to just have that in my pocket anywhere that I go that I can get that just in time coaching. So and it's a bit of assessment. It gives me good feedback. But I I I think we're in gonna be in a new era where there's all sorts of new assessment mechanisms that this AI tools will be able to give us.
speaker-1: Yeah, it's so interesting. I I mean I think one one word that comes to mind is intentionality, right? That that districts at the at the appropriate grade levels really need to be thinking about how to be intentional in the way that they're educating students about AI and the way that students are are interacting with the AI. But but we shouldn't I I mean I I think maybe a larger point is that we shouldn't kind of throw the baby out with the bathwater, right? We we the the the concern is real, but I think there's all kinds of, you know, and I I've been hearing myself all kinds of very interesting Use cases of increasing accessibility through AI for people that learn differently. The multimodal thing right now, maybe a student is an auditory learner, a visual learner. And the ability to, you know, as long as they're not kind of just getting the easy button to get the answer, the ability to have kind of a dialogue and and to seize these issues from different different directions can be absolutely incredible.
speaker-0: It can be, yeah. And I mean there's all sorts of ways now to engage with the material and to be and to be quizzed on it. aga again, you could just ask the voice mode and any of these tools to to just do a just in time quiz and it'll say something and it'll quiz you on it and it'll it'll give you coaching based on not just did you get it right, but how did you explain it? Did you stutter? Did you use ums and hus and again that's like a pretty powerful capability right in your pocket. I also you said something really important too. You said we shouldn't throw the baby out with the bathwater and I I I think that is another temptation right now is to think that AI changes everything. And it means everything inside of education needs to change too. We need new AI literacy and new prompting classes. I think that some of that is true on the margins, but in some ways it makes the fundamentals even more important. And you said before, and I I've said this too, that writing is thinking. And and that is particularly true as it relates to using these AI tools, that the the capabilities only get unlocked based on the sophistication of the prompt that you give it. And the and prompts aren't magical, they're they're English language written in a very logical way, in a coherent way, in an articulate way. And so, in many respects, it makes all the work we've been doing on ELA. even more important because it's the students who are proficient in ELA and in English and writing coherently and logically and concisely, they're the ones that are going to be able to unlock the prompts and the capabilities of these AI tools. And so in a weird way, the programming layer of this new era of AI isn't just math and isn't computer programming. It's also ELA. And so we should be leaning more into that now to help make sure that the this generation of students are prepared for the workforce they'll be entering.
speaker-1: Yeah. I love I I've always been struck and and really liked the concept that Ethan Mollock kind of came out with of co-intelligence, right? as kind of thing about the the way of thinking about how we're gonna interact and and and but it is you know it seems to me that we're this moment where, you know, kind of the technology and the pedagogy are are kind of, you know, they need to keep pace with each other. And then this goes right back to kind of a we started, like kind of a diffusion. story, right? How what's the r you know, it kind of feels like we're that's what we're grappling with most is how do we get the this incredible technology appropriately into the into the hands of educators and and ultimately into students.
speaker-0: No, I think that's right. It's like thinking appropriate. And also, you know, back we were talking before about different waves of technology. When the internet first came out, every district created an acceptable use policy. And this was this was very much about empowering local schools with their local communities to figure out what was acceptable, what was the right way of using the internet, using resources found on the internet, of using Wikipedia. There was a lot of concerns similar back then too, that students were just cut what they cut and paste what they found in Wikipedia into their papers. And so acceptable use policies allow communities to really kind of pace themselves with the technology in a way that's like values aligned and also aligned with their their parents and with their students. I think we need something similar to that right now. it's also you know, it's it's thinking about there's different types of AI. Like you know, there's these relational chatbots call it character AI, has no place inside the classroom. And I think you know, those. Those you can be pretty comfortable with banning. I think you have some of these general AI tools, ChatGPT, Gemini, you have you know, Claude, obviously. And they're, you know, you may not want to just completely give every single student access to that. Maybe you want to give your teachers access to that because it could be a powerful teaching tool. Maybe you want to hold that back from students or only allow students to use it in certain situations. And then you have purpose-built AI tools, and that is everything from PowerSchool to You know, brisk teaching, there's just a lot of these new AI tools that are coming out that have education safeguards built into it that have been designed with teachers and students in mind. And there, I think that that could be you can lean into that with a little bit more permissiveness. But also I think you have to be very clear on the problem you're trying to solve and and what does quality look like? How do we know something is improved by using that tool? And that is gonna be a a decision that really comes down to principals and to teachers to be. very clear about that and make sure that they're measuring the outputs and the outcomes.
speaker-1: And that brings our back to data, right? Like w ha you know, we need the data to actually tell us you know, ha what's what's the impact of these of these tools. interestingly, total aside, that we just did a a survey of over 350 district and school leaders for kind of the pack to school season, specifically asking like how satisfied they are with their ability to measure the impact technology.
speaker-0: yeah.
speaker-1: and and the news isn't great, right? Like very low satisfaction level. Yeah. really a lot of just using anecdotal feedback, very, very low in terms of actually being able to use kind of, you know, true efficacy feedback loops. And I think that goes down to like, you know, the A, there's a, you know, cause and effect is really hard to prove when there's a lot of of other factors, but also just the data that they might have available to them. So definitely some some work to do in terms of how do we give district and school leaders like better tools to understand the impact of these tools and how can we kind of accelerate the feedback loops on some of those things.
speaker-0: No, I agree. There's just one point on that, because I think it's it's it's an important point, it's a powerful point. But we've spent decades trying to give educators and principals and superintendents more data. We've done it through dashboards and we've done it with assessment data. And and I mean, in some respects, school systems and classrooms are swimming in data. And it the hard part has been making sense of it and seeing pattern recognition, seeing the coherence kind of within it and doing something with it. And again, maybe that is the killer app for AI right now. Maybe it has none. Nothing to do student facing. Maybe it is just helping teachers and superintendents and principals make sense of this data that they're they have already. And in if that is all we do, that could be immensely, immensely powerful. But but again, I think it's like it's trying to think through the tools and think through what is it that we're trying to improve and what is the best ways of collecting the data or analyzing the data to see if the patterns are there to show that we're improving.
speaker-1: Well y you make a great point. And it's a great it's a great kind of pivot point for us, right? Like so much of that energy in the focus and and understandably so for kind of AI in K twelve has been kind of the student facing the the the personalized tutor. but we're not thinking maybe as much. generally speaking on the impact for administrators, for teachers at a as a productivity tool, as an insights tool, and maybe ultimately as a or or eventually as an automation tool when we start thinking about agentic. I'd love you, we you met you kind of passed through very quickly a few use cases earlier in our conversation. Love if you come back to those, maybe talk about a couple where you're seeing some some interesting or the the the the highest potential right now.
speaker-0: I have struggled for last three years to try to figure out like what is the right metaphor for explaining what AI is. Is it a tool? Is it a service? And then I I I was with a chief technology officer out in Silicon Valley, and he had this great, he talked about AI giving people superpowers. And I was like, you know what? That actually is pretty accurate and pretty clever. And and I want every single teacher to have these superpowers because already we're asking great teachers to do almost the near impossible. They're we're supposed to go into a class of 25, try to understand where all those students Students are, where they're struggling, try to develop some sort of coherence amongst all the different applications and supplemental and core curriculum and all these other materials they have. They're trying to differentiate instruction. They're trying to make a sense of why some students might be chronically absent. We're asking them to do a Herculean job. And all of a sudden, now we might be able to give them some AI superpowers to help make some of that work easier, to help free up some of their time, whether it's on administrative tasks. Or to give them deeper insights. And and so I what I want is for every teacher to have those superpowers. However, they need to deploy them inside the classroom. Some of it might be developing instructional materials. We have a lot of teachers that are teaching out of field. They're amazing teachers, but they just don't have necessarily the depth and content knowledge. And so maybe what they need is a agentic teaching assistant that helps them get up to speed on some of the content before they go into the classroom. Maybe it's new types of assessment. Gentec AI is helping to just give a sense of where all the students are as they're coming into the classroom that day. So I just I think this idea of getting giving teachers superpowers. And once teachers have superpowers like that, they never want to give it up. And and I think that is the way to sort of start, I think with pilots. I think that's the way to start building some of the early AI use cases. But also I think this is perfect too on the administrative side. Like again, not even doing anything student-facing all of a sudden now. Principals and superintendents have an immense powerful AI tools that can help them with reaching parents, with better parent communication, with language translation, with helping to develop coherence amongst their schedules. There's a whole bunch of different use cases there just to help make school districts operate and function better for their students, for their teachers, and for their parents.
speaker-1: You have thousands of administrators and teachers vigorously shaking their heads up and down right now, nod nodding in agreement, particularly on the workload challenge. And and that's a that's a big focus for us for sure, is thinking about great, how do we, you know, what are some applications where we can kind of bring and you you hit on a couple which are great. What what are your, you know, how do you think about kind of the underlying kind of the the infrastructure requirements or are some of the ways that we can kind of you know. A AI, of course, is so dependent on kind of context and data and grounding. and so I love your thoughts on some of the ways that that, you know, districts can kind of think about setting themselves up for success. The way that maybe a lot of maybe enterprises that are already on the AI wave have been thinking the last couple of years and maybe have the flexibility to move, you know, at at whatever pace they're able to move to kind of put some of the grounding in place to to become more AI ready, more agentic, if you will, and and how how do you see that playing out in the district?
speaker-0: No, I think it's an important point you raised because these new models, it's been pretty much the case since ChatGPT four, but I mean, especially with these new models, the way you unlock so much of the capabilities is by giving them a lot of context. And it's not just giving them a better prompt or better direction, even though that's important. Prompts again are about defining what it is you want the AI to do, giving it guidelines about what good and what bad doesn't look like what good looks like and also what bad. bad outcomes would look like. And then a lot of context. And what I have found is that these these models are insatiable for context. And so if you think about it, either the context of your classroom as a teacher, you know, you have context with your school district, with, with school district policies, with school district curriculum. You have some flexibility. You've chosen a bunch of tools yourself. You're using different types of supplemental resources. you have a different set of lessons that you've done in the past and you have lessons that you have planning come out. All that is context. And the more context you give it, usually the better the output is on the other side of it. And I I think school districts and and schools can also lean into this too by helping it create skills and and other kinds of data sets that help the teachers AI agents and the teachers AI prompts with you know sort of pre-loaded set of context that they can just sort of plug and play inside of it. But I think context is going to be immensely important going forward and articulating that and getting that in a pretty rigorous way is going to help us make sure there's higher quality outputs on the other side.
speaker-1: Yeah. Yeah. And we were talking earlier, you were kind of mentioning this idea that we're in this you know, kind of inflection moment of going from AI as as a kind of a thing that you talk with. We're we're all kind of familiar with the chat GPT kind of chatbot experience, to a thing that that is taking action, that is doing things, more of the agentic model. Maybe you could talk a little bit about that and kind of give you give your view on that for for kind of the folks in our audience.
speaker-0: Yeah, I think it's it's for folks listening to this, it's it's you have to think of AI as like in sort of two different categories. There's the AI that I think most people have been using right now, and that's ChatGPT and Gemini and Claude, and it's it's almost like a super Google. Like they're using it to give to get an answer. Maybe you go back and forth with it a couple different times. And the task is like writing something for the most part, maybe generating an image, maybe generating a spreadsheet, but that's pretty much it. The era that we're moving into, and it is so important to understand because of how seismic of a shift is, is agents are very different. Agents are AI systems and models that act somewhat autonomously. So you give it a goal, and then it reasons on the best way of executing that goal. And it can work over longer horizons, and it can use more tools, and it can produce more things. And so Some of the AI agents that I've used have worked for 30 hours at a time on a problem. it'll build things, it'll build spreadsheets, it'll build just-in-time applications that you can use you know, for your work. Or I can imagine in the future, you know, teachers are gonna task an AI agent something and it'll develop. new resources for them to use inside the classroom as well. But agents are very different because you're prompting it, but you almost are acting like a manager. A agents are showing up in the workforce, in the healthcare sector, almost as employees. And so what you are doing is you're managing a team of AI agents and you're managing a team of humans. And that means you have to be very clear in terms of the task that you're assigning it. You have to be very good at reviewing the output of it and giving feedback and saying we're at mid met the mark and where it missed the mark. but you're managing and it's a very different type of skill set. but to give you an example, OpenAI just announced last week that they had a group of 10,000 agents solve a mathematical challenge that had that humans have not been able to solve for a hundred years. And what you have to understand is that OpenAI, yes, has access to enormous amounts of compute and sophisticated models. They had 10,000 agents working together to solve this problem. I think in two years, you're gonna have teachers that have the capabilities of tasking hundreds, if not thousands, of agents to work on a problem inside their classroom. That could be helping them with developing new resources. It could be helping them with figuring out what is the best way to help John understand this calculus problem since he seems really stuck on it. But I I think it's gonna be an amazing amount of capability boosters for teachers, but it's also very different because now you're not just prompting, you're also managing. And a lot of the ways these Prompting and giving a task to an agent looks is almost like a job description. Or it's almost like for a school superintendent, they could appreciate this. It's a scope of work. It's it's outlining what you want the agent to do, what success looks like, how you'll evaluate its quality, and then giving the agent the time and space and context to to actually work on the problem and come back to you to read.
speaker-1: view it. Yeah, that's amazing. And I can just already think of you know a half dozen you know, really important domains within a district where this kind of agentic approach could just kind of really alleviate so much kind of administrative burden on superintendents and on the cabinet, whether it's thinking about, you know, early warning indications for attendance, whether it's identifying, you know, an intervention earlier for a particular student or group of students. So really, really incredible to kind of think about where that where that goes. And and it has nothing to do with, you know, necessarily like AI, student facing AI in the classroom, it's just kind of a really thinking about how a district really transforms the way it operates. you know, use use
speaker-0: No, I think that's right. It's also thinking, you know, you almost you have there there's a lot of risks of anthemop anthromorpizing some of this too. But like, but but you almost want to think about like who is the team you want working on a problem. And so like I I think right now a lot of people are like, how do I get one of these models to be a really good tutor? And I think that where this is evolving is that you're gonna have a group of agents working collectively as a tutor. So you're gonna have a productive struggle agent. Whose whole job is to make sure that the student is actually engaged in a productive struggle. You're gonna have a coherence agent that's gonna make sure whatever the output is is all aligned with the district curriculum, supplemental curriculum, with the standards of the state. And you're gonna have a student engagement agent that really knows the student really well and is gonna say, okay, the best way of teaching John is by using football examples. And they're gonna work together as a team to help deliver. the experience for that student or for that teach that teacher. But I think it's a very different way of sort of thinking about things. I know it sounds like science fiction right now, but I'm seeing examples of this in healthcare. Definitely seeing examples of this showing up in the corporate enterprise. And so I think probably in the next year you're gonna start seeing agents showing up much more inside the classroom. And it's gonna give teachers a new set of superpowers as part of their portfolio.
speaker-1: So how do we think? Let's go back to a little bit where we started. We have, you know, and and we totally understand where where these are coming from. We have major districts that are banning AI tools at various grade levels. And there is a real risk that, you know, in the the again, the very real concern that district and and families may have about kind of that the the AI taking over, that the experiment implementation that's going to lead us to to those approaches, you know, might might be missed. I guess like how do we help, where's the balance? How do we help kind of you know provide the leadership for for districts to kind of see their way through this.
speaker-0: I think well, districts need to look at this as they can only move at the permission of trust. And the most important trust is gonna be with their parents. And I I think it's recognizing that parents are very anxious right now. They're anxious about again what's just happened with social media and with cell phones. They're anxious, I think, coming out of COVID, where they saw a lot of tech and not a lot of learning. And they saw a lot of harm. They saw a lot of their students. And we see it in the national education data with the PISA data. We we have just seen a generation of students that are behind. And and And parents rightly or wrongly are blaming a lot of that on the tech experience that happened. And so I think it just
speaker-1: Here comes AI, right.
speaker-0: like Lucy and the football from the peanut in the the comic strips where, you know, every single time they go to Charlie goes to kick the football, Lucy pulls the football away. And I think a lot of teachers, a lot of parents have said, we've heard this before. And so I I I I I think it's meeting them where they are and acknowledging that. I think it's setting up pilots to build that trust. I think it's also realizing that again, parents are hearing a lot of concerns about AI in the economy. They're hearing about AI disrupting jobs. And so, you know, they're trying to make sense of this in society, much less like deploying this inside the classroom. So I I would start first with trying to build trust with parents. I would start with very small focused pilots. And again, you have to be very specific and explicit with the problem you're trying to solve and what does good look like on the other side of it. and then also I I think there's a lot of room for districts to experiment with again, nothing that's student facing, but instead with just looking at some of their administrative and operations thing and how do they improve just the functioning aspect of of school systems. But I would sort of start there and and build trust as as they go. And just and also lastly, I would say be very explicit about what AI is not gonna do. Like I I think handing over decisions about grading. handing decisions over as whether or not a student su should progress to another class or to the next grade level over to AI. Those feel very high risk. And and so I would always make sure humans are in the loop, teachers are in the loop as part of making those decisions, and where AI is a tool and a supplement to it, not actually driving the decision.
speaker-1: Yeah. Moving at the speed of trust. That's a great that's a great line. that that it's the only way, right? Yeah, that we can do it. And and and I think, you know, honestly and and and I and I'm confident like the the promise is too great, like the intelligence is too amazing that we will eventually get there and and and some of the the vision that you're painting in terms of the this kind of team and and how it's gonna be applied to to really help a student be at the the highest point of their kind of contribution. I know we'll we'll get there in the end, but I I'm sure it'll be in some fits and starts. But if we do nothing else other than alleviate some of the burden on administrators, on teachers, give them capacity kind of back in their day to, you know, in the case of administrators, spend more time with a teacher, in the case of a teacher spend more time with the students. I mean, that will be in and of itself just a huge, a huge, huge win. Yeah. well, amazing. Love, you know, kinda s take a step back. Like what and and you might have already answered a little bit, but maybe maybe go further, like you know, five years from now, we we see this vision. What will what will district leaders have have done right to to get us there? Right. What what are the what are the steps?
speaker-0: I think if we have done this right, I think district leaders will have started with the first thing we need to do is build trust with teachers and with parents. because trust, you can't do anything. I think the other thing we'll have gotten right this time that we missed in previous waves is that we were explicit in the goals. We were explicit in what we were trying to improve. and we worked backwards with what are the best ways of using AI to test to see if we can improve the outcomes or capabilities on those. But if we're not ruthlessly focused on the goal and the problem that we're trying to solve, I what I worry about in some five years, we'll have the opposite. We'll have we have a lot of AI tools in the classroom. We'll be judging how many students are using it, how many lesson plans it's generated. And we'll have missed that it doesn't really matter. What matters is whether or not student learning increased, whether student engagement improved, and whether teachers had more hours that they could reinvest back in relationships and with giving feedback in their students. And so, but I think we have to keep in a laser-like focus on improving outcomes and having really clear measurable goals and problems that we're trying to solve.
speaker-1: Yeah, that makes a ton of sense. Let me shift gears for a minute. We've talked a lot about kind of application of of AI in in schools and different different use cases, but let's talk about kind of the the broader impact kind of from a workforce perspective. And I don't know if you have some thoughts on you know, current moment as it relates to how we provide how we generate a kind of a an AI ready workforce, right? Like it's it's, you know, we're feeling it in in our professional lives every day, but it's kind of changing the nature of work. And so curious if you have some thoughts on on how, you know, districts should be thinking about preparing students, how the system should be thinking about you know, those those types of outcomes for for students.
speaker-0: yeah, it's a great question. I I and first of all, there is such a rigorous debate right now about AI's impact and its disruptive impact on the workforce. And I think everyone's trying to make sense of it. It is hard because the models keep changing so rapidly, and trying to gauge whether or not like a ChatGPT five or you know. Gemini 2.5 has an impact on certain tasks inside the workforce. It's very different than as the agents are showing up and are capable of working longer and doing a greater range of different tasks. So everyone is trying to make sense of this. I I think the future is definitely ingenic. And what that means is it makes K-12 education even more important. And it and for the basics that we were just talking about, that it is impossible to manage a team of agents. if you can't clearly articulate what you want those agents to do. And so ELA is become the new programming layer and is more important now than it probably ever has been in the past. It makes math very important, not from a coding standpoint, because math teaches you how to think and how to reason and how to work through logic. And often the way you manage a team of agents is thinking logically and evaluating the output that they give you. And so math and reading become more important, not less important in this world. it also means every new entry-level job that's created, I think, in the future becomes higher skilled. And it's just because Capabilities of these AI tools, these AI models, and agentic AI means it can keep doing lower and lower skilled. It's going to eventually become higher skilled, which means every new job is gonna become higher skilled and needs a more skilled workforce for it. So K-12 education is even more important now than it has been in the past. but we have to be making sure that they're graduating proficient in reading, writing. in a proficient in math because those are gonna be the core skills to unlock this agentic future.
speaker-1: Yeah. It's so well said and and it kind of goes back to this idea like this this renewed focus on on learning how to learn and learning how to think as being kind of you know the critical skills in addition to you know all of the skill trades that we're gonna now need that out of kind of we've got resurged in in in in importance in our in our economy.
speaker-0: Trades have always been important. They're even more important now. you know, I also I I suspect there there's a economic theory called Jevron's paradox, which is when you make something cheaper, demand for it doesn't go away, demand goes up. And I think we're seeing that in a number of different fields right now where yes, AI is eliminating some computer programming professions, but we're actually seeing a rise in the increase in the number of computer engineers and computer programmers. And it's because all of a sudden non-coders like myself are writing programs and writing solutions to problems I have. But at some point you hit a wall. And who do you call when you you hit that wall? You need an engineer, but that engineer has to be higher skilled. Yeah. And so in many respects, I think we're gonna lower the floor because of AI ha giving individuals, teachers, students, and the average American worker access to this expertise, they could do more work they could never do before because it's now all of a sudden cost effective and they have a a team of agents that are available to them to do it. But that's gonna drive demand, I think, for a lot of these other professions. And so Yeah, I I'm a little bit more optimistic on this. I I I think there's a lot of folks who are concerned about AI is gonna massively disrupt the workforce. I think there's another scenario here where actually we have labor shortages. And and again, all that comes back to education. And we have to make sure students are leaving high school with the fundamentals to prepare for any of any of those scenarios in the world.
speaker-1: Yeah, absolutely. Well, last question. you have the opportunity to give a K twelve district leader, a superintendent, advice on how they can build a stronger school today. what would what what's your advice?
speaker-0: I would my advice would be lean into this AI moment with a lot of curiosity, lean with pilots, be slow to scale, keep a high bar for scaling, but lead with pilots. I no one is f no one has this all figured out. And the only way to figure it out is by trying it. You'll find use cases with your teachers and your community. that others won't. And the moment you find those use cases, it unlocks that superpowers. But I you have to experiment with it. I I there's a great, there's a great team called Decoding Futures up in New York, and they they talk about learning AI is a little bit like learning a foreign language. There's only so much you can learn just by reading about it. You have to practice it. And so, what I would encourage is superintendents and principals to really start practicing with it. Just try it. If you had, what would you give a remote worker as a task? Just try that. And and whatever that email is that you would write to that remote worker, use that as your prompt. That begins unlocking use cases. and then the moment you start unlocking one, you start seeing a hundred. But I I think this is a remarkable moment for superintendents and for principals because they get this immense capabilities. that are coming in the form of AI and agentic AI in the future. And it's gonna help them unlock a whole nother generation of teaching and learning that I I I think can lead us into an amazing set of prosperity. But again, it's gonna be up to the leaders to harness it and to use it in a strategic way to to make that happen.
speaker-1: Well, on that note of abundance and positivity. John, thank you very much for joining Stronger School. Really appreciate you joining.
speaker-0: Thanks for having me. Thanks for running.
speaker-1: Yeah, you bet. All right. Thanks everyone.
