Edtech Insiders
Edtech Insiders
Teaching Lab Studio: Co-Designing AI Tools with Educators
Dr. Sarah Johnson is the CEO and President of Teaching Lab and Relay Graduate School of Education, leading their AI-enabled product innovation and educator preparation initiatives. She is joined by Teaching Lab Studio fellows: Riz Malik, creator of Coteach, a curriculum-aligned AI assistant for math teachers; Gautam Thapar, CEO of Enlighten AI, a personalized AI grading and feedback platform; and Louisa Rosenheck, co-lead of NISA and the Tangle & Thrive research project, focused on AI-powered instructional coaching and student engagement.
💡 5 Things You’ll Learn in This Episode
- Co-designing AI with teachers
- Curriculum-informed AI boosts quality
- AI scales coaching & grading
- AI-centric classrooms & tool ecosystems
- Teaching Lab–Relay expands R&D
✨ Episode Highlights
[00:03:12] Sarah on launching Teaching Lab’s AI studio
[00:07:47] Riz on curriculum-aligned AI for math
[00:10:50] Gautam on personalized grading AI
[00:16:44] Louisa on AI-supported coaching
[00:24:40] Educators building GPTs
[00:41:05] Spotlight on Studio projects:
- Mathly by Melanie Kong & Marie White
- Writing Pathway by Sherry Lewkowicz & Phil Weinberg
- Podsie by Joshua Ling
😎 Stay updated with Edtech Insiders!
Follow us on our podcast, newsletter & LinkedIn here.
🎉 Presenting Sponsor/s:
Every year, K-12 districts and higher ed institutions spend over half a trillion dollars—but most sales teams miss the signals. Starbridge tracks early signs like board minutes, budget drafts, and strategic plans, then helps you turn them into personalized outreach—fast. Win the deal before it hits the R
This season of Edtech Insiders is brought to you by Cooley LLP. Cooley is the go-to law firm for education and edtech innovators, offering industry-informed counsel across the 'pre-K to gray' spectrum. With a multidisciplinary approach and a powerful edtech ecosystem, Cooley helps shape the future of education.
Innovation in preK to gray learning is powered by exceptional people. For over 15 years, EdTech companies of all sizes and stages have trusted HireEducation to find the talent that drives impact. When specific skills and experiences are mission-critical, HireEducation is a partner that delivers. Offering permanent, fractional, and executive recruitment, HireEducation knows the go-to-market talent you need. Learn more at HireEdu.com.
As a tech-first company, Tuck Advisors has developed a suite of proprietary tools to serve its clients better. Tuck was the first firm in the world to launch a custom GPT around M&A. If you haven’t already, try our proprietary M&A Analyzer, which assesses fit between your company and a specific buyer. To explore this free tool and the rest of our technology, visit tuckadvisors.com.
[00:00:00] Sarah Johnson: I think that is the primary value proposition for entrepreneurs that we recruit to Teaching Lab, which is as we scale, we have more and more test beds for innovation. More and more opportunities to do user-centered design, both in our coaching partnerships and now with Relay who leads the highest impact teacher prep program in the nation in our teacher prep partnerships.
And then additionally, together we'll have one of the largest leadership development partnerships. So you can imagine that over time. More and more people that we bring into the studio can access novice teachers. Teachers who are teaching right now, leaders, and both learn from their experience and two test tools with them and co-develop with them.
[00:00:46] Alex Sarlin: Welcome to EdTech Insiders, the top podcast covering the education technology industry from funding rounds to impact to AI developments across early childhood K 12. Higher ed and work. You'll find it all
here at EdTech Insiders. Remember to subscribe to the pod, check out our newsletter, and also our event calendar.
And to go deeper, check out EdTech Insiders Plus where you can get premium content access to our WhatsApp channel, early access to events and back channel insights from Alex and Ben. Hope you enjoyed today's pod.
Today we're talking to a panel. Of amazing fellows from Teaching Lab Studio as well as Dr. Sarah Johnson, the CEO and President of Teaching Lab and Relay Graduate School of Education. Dr. Sarah Johnson led Teaching Lab's launch of AI enabled education tools in 2024 and is overseeing the integration of Relay and Teaching Lab.
Dr. Johnson previously led initiatives at the Overdeck Family Foundation and the New York City Department of Ed and has a doctorate from Harvard. Rizwaan Malik leads Coteach a curriculum aligned AI assistant that helps teachers with curriculum implementation and impactful classroom practice. He's also affiliated with Stanford's EDUNLP lab, studying how AI can advance teacher practice.
Riz has a background in research and data science and began his career as a math teacher in the uk. Gautam Thapar is the co-founder and CEO of Enlighten AI, a personalized AI grading assistant that helps educators deliver better, faster feedback supported by New Schools' Venture Fund. He previously founded Invictus Academy, a top performing charter school, and began his career with Teach for America.
Louisa Rosenheck Co-leads NISA, an AI powered instructional coaching platform, and tangle and thrive, a research project around engagement and motivation as a learning designer focused on playful inclusive learning. She has worked across research and industry from MIT to Kahoot, and she teaches Ed Tech design at the Harvard Graduate School of Education.
Dr. Sarah Johnson. Riz Malik, Gautam Thapar, Louisa Rosenheck, welcome to EdTech Insiders. It's great to be here. Thanks for having us.
[00:03:12] Sarah Johnson: Thanks, Alex. Thanks as always.
[00:03:14] Alex Sarlin: I'm so excited to talk to this group. I think Teaching Lab Studio is doing some of the most interesting and innovative work in EdTech, and I have followed it for a long time.
But I imagine many of our listeners may not know everything about what you're doing. So first, let me kick it off with you, Sarah. Tell us. About the Teaching Lab studio, how did it start and what is its goal and what makes it really unique and special in the ed tech space?
[00:03:38] Sarah Johnson: Yeah, so I think we have the most unique ed tech studio in education, and that is because I am a trained educator.
I was a high school science teacher and for years, teaching lab was only a teacher coaching organization. So we are an organization of educators. And what we were realizing is there were all these ed tech developers coming to us and asking us to get teachers to use their tools. But the problem was if they were coming to us to try to get teachers to use their tools, they were already a little too late.
Hmm. So what we realized is they weren't actually developing the tools in the real context of classrooms, the real workflows of students and teachers. And with the advent of ai, we realized maybe we could do this ourselves. We started a studio, even though some people thought it was a crazy idea for a service organization to stand up a product development side of their organization, and we started an AI and education fellowship.
We had over 800 applicants for that fellowship in the first two weeks. And you are talking to some of the amazing fellows that we recruited during that time period. And what the fellows are tasked to do is come up with use cases that are aligned with exactly the pain points of teachers and students work with those teachers and students.
And Riz can say a lot more about this, about like all the r and d that he did before he developed Coteach and then. Once they design something, iterate on it with teachers and students to ensure that we achieve two goals. One, obviously that we're having an impact either in improving teacher practice or student learning.
And two, to ensure that teachers and students actually want to use the tools. So there's another big problem in education, which is called the 5% problem, and that refers to the fact that even when there's ed tech tools that could be really effective, it's only 5% of students that use them at the dosage required to get that impact.
So you need to do just as much r and d on impact as you have to do on usability.
[00:05:39] Alex Sarlin: You're mentioning impact and usability. But one thing that is not the absolute top line of teaching Lab Studio, which I think also makes it really unique and special, is revenue. This is a studio that's really designed to create ed tech products that work and that people love, and the revenue side is not paramount.
Tell us a little bit more about that model. 'cause I think that is something that sadly, a lot of other studios and venture models don't get to prioritize.
[00:06:05] Sarah Johnson: That's correct. Yeah, so that is built into the DNA of the studio. So there's a big problem in education, which is there's philanthropy for some r and d, and then there's what some people refer to as the Valley of Death.
In between that r and d and actually. Scaling products. So we think the studio sits within this valley of death and tries to bring life, perhaps to that valley, and we want to give entrepreneurs the runway they need to actually ensure that the product can have an impact before they try to go and sell it.
Because what we see out there in the market is that unfortunately, some of the most dominant. AI tools in used by teachers and students are not the most effective. So the market incentives are all kind of whack, which is why we need more people, more organizations, more investors, to actually help us solve that valley of death problem.
[00:07:01] Alex Sarlin: That's a really great point, right? I mean, scale is something people just prioritize right up front, but you're saying get it right. Make it something people actually love and that actually works in the classroom before you start marketing. The heck out of it and trying to get it into as many classrooms as possible, and that, unfortunately, is often not how Ed Tech works.
Let's go around the room and talk to each of these fellows about some of what they're doing, because the projects that come out of this model, this is part of what I really love about Teaching Lab Studio, the projects that come out look and feel pretty different and really a lot more innovative in many cases than what you see from a purely market driven model.
So Riz, your name was invoked. Let's start with you. Tell us about what Coteach is. Some of the r and d that you did before you came to Teaching Lab to prove it out and make sense of it and what it's like working in this type of teaching lab studio environment.
[00:07:47] Riz Malik: Yeah, of course. So Coteach is a curriculum aligned AI assistant that's been designed specifically for math teachers at the moment.
And so we think of Coteachers, like the bridge between high quality instructional materials and then the day-to-day realities of what's happening in classrooms. And so teachers use Coteach to tweak their lessons, to create scaffolds, create more practice, or even just like to understand what's already in the curriculum.
Sarah kinda mentioned some of the r and d work that went into all this, so I've been pretty obsessively interested in curriculum and lesson planning ever since I started my career as a middle and high school math teacher. And then about two years ago I was at Stanford doing some research. We were looking at how teachers actually use.
High quality materials in their classrooms. We were watching them on Zoom planning lessons and then interviewing them, and a couple things really stood out to us. So one is that, I mean, these materials, obviously, they're very, very important. They are very helpful for particularly early career teachers who are still building their own pedagogical content knowledge and everything else.
But they're also difficult, like they're dense, they're very long documents, they're hard to pass. There's so much already in there. And what we observed was teacher was spending a lot of time taking these materials and then tweaking them, kinda tailoring them, adapting them, trying to understand them for their particular classroom.
And that was tough work. It was really hard work. And so kinda when I joined the fellowship, uh, teaching lab, all of the work I've been doing so far has been thinking about how can we build out that bridge between the materials. And then what actually needs to happen on a day-to-day basis to make the curriculum work for students.
And so when I first kind of came on board just over a year ago, we kinda cycled through a bunch of different prototypes. We had a Chrome extension at one point, we had another prototype. And then early this year we settled on a very simple idea, which was to build out an AI agent that just had complete access to the underlying curriculum.
And so we built that, we launched it. It's aligned to Illustrative Math in March. Then we've seen very, very quick traction with that model, and so a lot of teachers are now are using Coteach and we're seeing a really fascinating range of use cases kinda emerging pretty organically.
[00:09:53] Alex Sarlin: It's really exciting and I think, you know, I'm nodding along and I imagine anybody listening to this who's worked with high quality instructional materials, like illustrative Math feels the same way.
They're incredibly high quality, and they're proven and they're evidence-based, and they're amazing. They're incredibly dense. There's so many, for every lesson, there's all these variations and these breakouts and just all these different pieces. The idea of using that as a corpus of material. The teachers can then build on and incorporate into their actual workflow is incredibly inspired.
I think it's a terrific idea. I know, Sarah, you were just on a panel at EdTech week with Kiddo. We've seen a whole sort of cottage industry form around these HQIM materials and how to get them to actually get the impact that everybody, everybody thinks they should have. And I feel like you're doing something really, really powerful in that.
Let me pass it to you, Gautam. We have talked a lot. I've worked with you in a number of different capacities, and I think what you're doing is fascinating, but for those who don't yet know what Enlighten AI does and how it does feedback, tell us what it is and again, what your experience with Teaching Lab Studio has been.
[00:10:50] Gautam Thapar: Yeah, absolutely. Enlighten AI aims to be your personal teacher's assistant for the purposes of grading feedback and data-driven instruction, especially for writing. And so if you think about what we would do, if we. Staffed every single teacher with a human teacher's assistant. There's a few different ways that they would augment, grading and feedback.
The first would be for summative tasks. You would sit with your TA after class, you'd hand 'em the rubric, the task, but you wouldn't just send them off and say, alright, I'm sure you're gonna do this really well. And just like, I want. You would train them with examples, you would actually coach them up and at a certain point they could take the stack, bring it back to you for a final review, but you get a lot of leverage off of your expertise, the more cognitive work that is more fun for teachers and more worthy of their time.
And so that's one of the things that we do. You can train. An AI assistant to actually grade and give feedback, but it learns from you in the process, and that keeps the context of the classroom, the context of your curriculum all front and center. After we did that, a lot of teachers were maybe more impressed by how the AI would reflect their voice and their standards.
Then we even thought possible at the time. And so they pushed us to say, well, if I really had a ta, what I would probably do is actually have them circulate class. And when students raise their hand, or when students submit a draft, oftentimes they need some coaching, they need some support, and I'd love for them to revise on the spot.
And so what we've done is we've flipped that script and so you can actually train the TA in advance of class. Send that TA out to basically give students on demand feedback and enable them to revise for those more formative assessments. And then the last piece of that puzzle was, well, if this TA is really, really good, maybe they're.
On their way to earning their teaching credential themselves, maybe they could draft the resources for the next day. So I can reteach based on the learning gaps because the data and writing in particular is just such a black box. I know this from when I was teaching in the classroom. The urge to just pump out feedback, get scores in, and just not that late, that often makes it difficult for us to use or leverage the data or really think about the data in a meaningful way.
And so enlightening AI also will sort of. Roll up the trends, the insights for each individual student as well as for the class, and you can actually develop a reteaching resource that you can use the next day to target those learning gaps. And so we want this formative assessment loop, especially for all written tasks, to be much more aligned with what the research says is best practice for kids, and also much more sustainable and scalable for teachers.
[00:13:18] Alex Sarlin: Yeah, it's an incredibly exciting idea. Just personalizing the teaching assistant in a classroom so that it actually reflects what a teacher truly wants to reflect in a classroom. It's not just an off the shelf auto grader that's gonna do it its own way. It learns from the instructor and then builds on that knowledge to create more and more value in the classroom.
Tell us a little bit about your teaching Lab studio experience. What has it been like taking this idea? In this really amazing collaborative environment of teaching lab studio where teaching and impact and usability take center stage.
[00:13:49] Gautam Thapar: Yeah. I really don't know if we could have built the tool in this way anywhere but the teaching lab studio.
And the reason for that is that grading and feedback is really hard. So unlike a lot of use cases for ai, there is a better answer. And in fact, sometimes a best answer when it comes to the score that you should give a student and the kind of feedback that you should leave a student. And so. Well, AI can produce something reasonable, and in some use cases, that's a helpful starting place that the teacher can iterate on.
There's bigger implications for something where there is sort of a right answer in many cases. And I think one of the things that's been really great about working within the teaching lab is we've been pushed to really think about that research. So I'll tell you one of the experiments that sort of worked on with the leadership team and that I found most interesting.
We basically took a stack of papers that had been graded for. A New York City exam that's on the New York State rubric, and it was already human scored. And we said, okay, if this had been graded with enlighten, how similar would those scores have been? And we actually measured this statistically. And then we were pushed even further to say, well, okay, great.
It seems like it's pretty accurate to what this teacher did, but how accurate is it relative to like somebody who would read for this state exam? Mm-hmm. So we went out and we recruited somebody who read for the state exam and we said, well, this is the best you're gonna do in the field, right? You're gonna have a seasoned administrator, somebody who actually reads for this test, and if you compare their scores and then compare them to the AI and really analyze that.
That should really tell you whether this is moving the needle in the way that we want. And what we found was really surprising, which is we expected the humans to at least slightly outperform the ai. And what we found was that by a factor of 25% to 50%, the AI was actually more tethered to the human. And even more interesting is it really depends who that human is and how you calibrate the AI does in fact make a difference.
So this, it puts language to something that we all know, which is the context that you provide the ai. Really matters for the quality of the output. And that's why it's been so cool to build in the Teaching Lab studio to see stuff like what Riz is doing and what Lisa and Issa are doing. Because we really think about like, how do you make sure that the output is actually high quality and not just sort of reasonable and kind of nice to have to speed us up, but to really make impact.
[00:16:04] Alex Sarlin: And reflect the priorities and the voice and just be really specific to the circumstance it's in. Whether it's administrative state grader or a teacher in a classroom giving formative feedback or summative feedback. The context matters so much, and I think it's something that often gets lost in the conversation about ai, where people think it's just taking something off the shelf or going to an open tool and using it.
And it's not. It's about working with it over time, teaching it how you think. It's really fascinating. So, Lisa, let's talk about NISA. It's an AI powered instructional coaching platform, and you're also working on a research project around engagement and motivation called Tangle and Thrive. Tell us about these projects and what your experience with the Teaching Lab Studio has been like.
[00:16:44] Louisa Rosenheck: Sure. Yeah. So NISA is an instructional coaching platform. The, I think the exciting thing about NISA is that it's really focusing on. Using AI to amplify connection and care. There's a lot of research about instructional coaching and how effective it is, but so much of why it's effective is the human relationship, the trust, and the time spent and the personal knowledge of the teacher.
But we also know that instructional coaching is very hard to scale because it's very resource intensive and schools don't have enough coaches for every teacher to have access to that. Resource. So that's where NISA comes in. NISA is really an ecosystem of a set of tools and it is meant to extend the reach of instructional coaches, so it does not replace them.
It is meant to fill in the times between visits and observations so they can deepen the relationship and provide additional kinds of supports and practice opportunities to teachers. But it's all based on. What the coach personally knows about the teacher. So based on their observations and their notes and the action steps they've identified that the teacher should work on and the specific curriculum and the context of their school.
So all of this goes into NISA, and then NISA can use all of this information along with. What it knows about evidence-based best practices in teaching, in coaching. Lots of things that we can talk about in a few minutes, but it puts all this together to give tailored suggestions and resources to improve teacher practice and then of course improve student outcomes.
So lots of really cool stuff going on on NISA, both around this human element and the technical components and what strategies are we using to make it feel human. So that's NISA. Happy to talk more about that. And then, yeah, we also have this exciting r and d project called T and Thrive, which is looking at.
Engagement and motivation first in the context of middle school math and digital curriculum platforms. But really that can be extrapolated more broadly. And this project is very connected to what Sarah mentioned about the 5% problem. So kind of looking at why is it that 95% of students are not accessing these tools the way.
They're meant to be or the way we want them to be, and what are the ways that we can nudge students and schools to spend more time on them, use them with more fidelity, but also to design those platforms to engage students more authentically and more deeply, which will then also support their use and their kind of minds on engagement with those tools.
And that project is a really neat one because. We're kind of experimenting with a variety of different approaches and then what we develop, what we design could potentially be attached or integrated with a number of different learning tools. So I think that's something that's really exciting about teaching Lab Studio.
Like Sarah said, the immediate goal is not productize, commercialized. Revenue, it's impact. So how can we really make tools better? And if that means creating something that is part of another tool, then that's what we should do. So that's a great project where we get to explore and think about how to improve ed tech in many different ways.
[00:20:06] Alex Sarlin: That opportunity for collaboration and sort of integration of tools within the studio is also, I think, something pretty unique. You don't often see that in other studios or in other formats where people can sort of work together or take their projects and borrow pieces and put them together to really add up the impact.
So, Sarah, as you hear all of these fellows describe their projects, of course you know them really well. Teaching Lab has also recently merged with the Relay Graduate School of Education, which is another amazing institution. Your organization is all about building these incredible important, I almost think of them as like last mile tools, like things that teachers are truly struggling with.
You know, incorporating HQIM properly and spending all the time on the PDFs, calibrating their grading, and having teaching assistants, getting instructional coaching that actually works. These are really important tools. How does it integrate with the Teaching Lab and the Relay Graduate School of Education?
How do you actually bring together the educators in your orbit with the fellows that we're talking to in this call?
[00:21:02] Sarah Johnson: Yeah, so actually I think that is the primary value proposition for entrepreneurs that we recruit to teaching Lab, which is as we scale, we have more and more test beds for innovation, more and more opportunities to do user-centered design, both in our coaching partnerships and now with Relay who leads the highest impact teacher prep program in the nation in our teacher prep partnerships, and then additionally together.
We'll have one of the largest leadership development partnerships, so you can imagine that over time, more and more people that we bring into the studio can access novice teachers. Teachers who are teaching right now, leaders, and both learn from their experience and two test tools with them and co-develop with them.
I think the next level innovation that we're pursuing at Teaching Lab and with the Relay integration is we're trying to create. AI centric classroom models that are not just about testing one tool, but are actually about knitting together all of those things. Right now, you cannot buy coherence. You cannot buy one platform, one tool that is going to understand everything that's going on with students.
Give feedback to teachers in the way that NSA is giving feedback to teachers, support teachers to plan for instruction. Like Coteach is supporting them. Give feedback to students based on like the context from the students and the teacher. Like Enlighten is, there's nothing that does all of that together.
So over time we wanna create r and d sites where we're testing those AI centric models and then actually work up to building AI centric schools. So that's the sort of next stage for the studio as we also plan for this partnership with Relay.
[00:22:43] Alex Sarlin: That's really exciting. So just a follow up question on that.
With Relay and with leadership development sort of coming into play, there's such a virtuous cycle that you're sort of talking about there. The co-design aspect where educators and innovators are working together to create really important things. Many of the innovators are also ex educators, many of them on this call, but when you put them together, it's sort of more than the sum of its parts.
I think one of the things that's also really interesting about this model is that. The teachers, the educators who come up through this model, through Relay in this format, are being exposed to AI and understanding it in such a deep way. They're trying all of these different tools. They're trying to put them together into coherent model.
The ones who start teaching in these AI centric schools will have this hugely different perspective on AI than those who have been in the classroom for 10 years and AI was sort of dropped on them by their administrator, and they're sort of not sure what to make of it. I'd love to hear you talk about that.
Do you envision that being part of the. Model as well is creating the next generation of AI enabled educators.
[00:23:41] Sarah Johnson: Absolutely, and I mean, I think you heard it in the story Riz was telling, and I put a note in the chat that we got from a teacher that really, I think, exemplifies what we're trying to do. So in the early days of creating Coteach, we convened teachers and we actually taught them how to create their own GPTs based on their pain points teaching illustrative math.
And we received a note from a teacher who was a part of that group and then. Started using Coteach later and then became a coach, and now is coaching teachers to use Coteach. And he wrote us a note about just how amazed he was by the progress, right? And how excited he was to be a part of that early user-centered design group.
And we've actually done similar things in literacy. So I think actually the best way to co-design with teachers is to teach them about how to use ai. We don't think that teachers are gonna wanna create a hundred GPTs on their own, but when you do that work with them, then they are way better consumers of tools in the future.
[00:24:40] Alex Sarlin: So Riz, one of the things that's incredibly interesting about what you're doing with Coteach, and I think it's a little bit of a trend right now, a really positive trend is what some people are calling curriculum informed ai, right? Grounding the AI in a curriculum, a particular curriculum, a particular set of high quality instructional materials, so that rather than using the giant corpus of material it was trained on, which is.
Everything, the entire internet, all of Reddit, it's actually drawing its material directly from a curriculum. Can you tell us about that approach and why you feel like that's so powerful for education?
[00:25:12] Riz Malik: Absolutely. So I think one of our earliest observations was that when teachers were trying to use the kinda very powerful, large, pre-trained models, the big challenge was that they just didn't know about plus of math or EO education or whatever it might be.
And so I think when we align to high quality curriculum materials, we're seeing two big benefits. The first one is around efficiency, so as we know, there's so much already in the curriculum, and navigating all of that is just really difficult. And so one of the things that we see teachers using Coteach for is they could ask a question like, where does this concept come up later on in the unit or later on in the grade level?
And they could go through every single PDF and find the answer themselves, or they could have an AI agent kind of run off and look through every single lesson and every single unit, and then come back with that answer. And so it makes it easier to navigate what's already in there. And then the other big piece then is just about quality.
So I think. The key thing is that when AI is in grounded in the pedagogical philosophy and approach that underpins a particular curriculum, so in our case, Coteach it understands the importance of low floor high seating tasks. It understands the importance of discourse in the classroom, all the things that are really important to illustrative math.
And so then when teachers are working with Coteach to do things, the responses from the AI respect the underlying philosophy and approach of the curriculum, and that makes for a much more coherent feeling for the teacher. And then the other thing that we've been observing is it actually means that the AI has to create and generate less.
In some cases, often the answer to a teacher's query or question actually already exists in the curriculum somewhere. And so if a teacher comes in and says, how could I support my multilingual learners on activity two? Well, the curriculum materials might actually say how you might wanna do that. And so we can already kind of retrieve that and then use that information when we come back to the teacher with a recommendation or a resource.
And so actually it means that there's less. Content being generated by the AI because it's kinda grounding in what's already been written by a teacher or a curriculum writer and what's kinda already in the evidence base. So yeah, I think efficiency and quality have the been the two big things we're seeing.
[00:27:17] Alex Sarlin: Yeah. It's interesting to hear you describe it. It's almost like there's a combination of AI and search functionality baked into it, because illustrative is so vast that the answer may be somewhere in there. Directly written by a stone called expert in math instruction, which is everybody at Illustrative, but you can't find it without going through 10 different documents and then piecing it together.
Quick follow up there. Do you see differentiation as also a key aspect to, are there teachers going in and saying, well, my classroom is half multi-language learners, or My classroom has this particular feature of it. How do I adapt it for that? And does Coteach help with that?
[00:27:50] Riz Malik: Yeah, I would say that it's probably the biggest use case we're seeing right now.
So I think whenever you talk to teachers across the country, the biggest pain point we are hearing is students are kinda working below grade level and then they're trying to kinda work with these on grade level materials. And so a huge area where people are using Coteach is asking questions like, how could I scaffold that activity for this group of students?
Or maybe we saw a great use case recently. A teacher kinda took a photo of a misconception that they saw in their classroom and gave it to Coteach and then asked Coteach to help them think about how to adjust the next lesson to kinda differentiate based on what they're seeing in the classroom already.
So differentiation, scaffolding are two huge areas of use at the moment, and I think that's great because I mean, it's helping teachers keep on the grade level content and then kinda think about the right supports and scaffolds that might be needed to kinda get them up to that point. And so. The alternative before was often teachers were not using the grade level materials because they just felt they weren't appropriate, and so that's kind of why we're working.
[00:28:50] Alex Sarlin: Yeah, and that use case of being able to sort of put the pieces together across days to sort of learn how a teacher is teaching, what their classroom is about, what kinds of adaptations they need, and then support them over time just adds up even more. And I think it reminds me of what you're talking about, got them with enlighten in that you're seeing these additional use cases.
For what started out as just summative feedback and then becomes formative feedback where students can actually check their work against an ai, which is a very popular use case for students. By the way. That is something that students really love doing is getting quick feedback from an AI about an assignment while they're working on it.
So I'm sure that's really popular. It has been across the board. Last thing you mentioned is it starts to build. Teaching material for the next day, and it's sort of going up the stack of what a teaching assistant would do. That's a really exciting way to look at it. And I'm curious, I'd love to hear you talk sort of broadly, because it's core to the teaching lab Studio, DNA is to support teachers where they're at.
You know, this idea of building AI tools that can climb the ladder and become a more and more. Valuable in different use cases sort of become really the right hand, you know, man or woman to the teacher in multiple types of teaching. It just feels like a very promising direction for how AI could truly make its way into the classroom.
It doesn't have to do everything on day one. It has to sort of make its way, earn the trust of the teacher and then do more and more.
[00:30:08] Gautam Thapar: Yeah, absolutely. I think as we have built out the first use case, which was what we thought. Would be the last use case, which is around the summative grading and feedback, which thought was a hard enough problem as it is.
We've just sort of tugged at that thread. And what we've arrived at, I think, which is not dissimilar to either of the other two products, um, that we've been talking about is there's a system that. Is in place, and there's a problem that that system is solving. And AI is enabling the system to exist where it wouldn't have otherwise.
And so even building on just what we described earlier, where you know, first it was summative grading, delayed feedback to students on more traditional assignments, then it was immediate feedback and revision for students. Then it was the resource after class. Probably the most interesting current workflow for us is now we've been asked by administrators, schools, districts.
How do we scale up the expertise of our content leaders and of our coaches, right? So if my teacher can train their personal ai, what about for our interim assessments where, you know, every teacher getting in a room, calibrating, scoring, consistently, getting all that data in, getting all that feedback out.
Like that is a huge pain point. When I was a a principal, we would shut down school for a couple of hours. Everybody from, you know, the front office manager to you know me, to you name the person on campus, they were scoring and giving feedback on essays. We almost never finished. When we looked back at the data, the scores were just all over the place, despite our best efforts to calibrate.
And then we ended up not really using the data in a meaningful way. And so what schools have been asking us to do is say, well, could I have my curriculum? Expert or my coach or my content lead, could they train up the AI almost training the teacher's assistant that that then goes around to each classroom in order to support the teacher with the teacher still being the final arbiter of what goes back to students, but just such a more seamless process that yields this consistent data, these x-rays of student performance.
And so I think one of the coolest things about this work has just been, as you tug at the thread and talk to users and talk to different stakeholders. We just uncover more and more different use cases that that really fit this system that we're trying to build that ultimately, you know, has a really big impact.
We're seeing just really exciting stories of schools that are moving the needle and writing assessment and ELA proficiency. And so at the end of the day, this is all having a student impact.
[00:32:32] Alex Sarlin: That's a very exciting vision. I love that idea of scaling expertise from different stakeholders within the education ecosystem.
It can be from a classroom teacher, it could be from a writing coach, it could be from a, a district administrator. I love the idea that, you know. Training an AI to act like a specific person and literally training it over time, not just telling it to act like a certain person, training it, working with it opens up all of these incredible possibilities for basically increasing the resources available to students at any given moment to teachers.
Uh, it's really exciting and I think. It dovetails really well with what you were saying about instructional coaching. Louisa, you were, you were mentioning something that I wanted to double click on here, which is that NISA builds on best practices. It knows instructional, you know, best practices. It knows the research and that is something that I think is an incredibly exciting aspect.
You have a background at, at MAT in Harvard. You have come, come from a long academic world and you know this research really, really well, and the idea of being able to take. You know, just as taking things out of the confines of HQIM or taking them out of a, the data out of a calibration session, that's usually hard to use.
Taking the knowledge out of the repository of research, everything we know about education and actually putting it to use at like the point of use feels like one of the most exciting use cases. It's something I think you're doing. I'd love to hear you talk more about it.
[00:33:49] Louisa Rosenheck: Yeah, so this is really our approach to context engineering in EdTech, and it is really exciting.
I think it's a really important method that really all AI tools should be using. And you know, Riz touched on it also in terms of having the AI know the curriculum. Everything from the curriculum is in there. So that is one type of context engineering, right? It's not just guessing at, oh, what I saw on the internet about.
A certain curriculum, it's, you know, everything is there. And similarly, you know, the AI should know the content and the curriculum well, it should know the pedagogy well. And pedagogy is always the thing that I'm pushing on. And the problem with, you know, foundation models is of course they're trained on the whole internet and.
What is most prevalent on the internet is often not what is evidence-based practices, right? Right. We know that what's just out there, what is most commonly done in schools is not necessarily what research says is most effective. So. You really cannot have an AI tool for EdTech that just draws from that.
You have to really intentionally curate the, you know, the research and the practices that your AI knows about and direct it to draw from that, prioritize that. And what we're finding is not only you just, you know, give it a bunch of research papers that's, that's not enough, uh, because it's too, it's too theoretical, right?
The, you still have to kind of translate, like tell the ai, okay, this is. This, these are the best practices. This is what the research says, and this is how it looks in the classroom. So this is something we've, we've worked on a lot in NISA we call it NSA's brain, or sometimes we call it the evidence-based backbone.
Like it's really what, what holds NISA up. And literally what it is, is, you know, some information on best practices. So like curated lists. And again, this is where we're also in a very unique position to be part of the, the larger teaching lab organization. Where it's full of expert educators. Expert coaches, yeah.
They know this stuff so well, and we're so fortunate to be able to reach out to them and draw on their expertise and then put that into NISA, so. We have, you know, not only this sort of curated resource of best teaching practices, but also we've worked with coaches to write up, you know, an AI readable kind of a format.
What does that look like in a middle school math classroom? What does it sound like? What should the teacher be saying? What, what would the students be saying if it's successful? What would a coach tell a teacher to help them implement this teaching move? That kind of thing is not necessarily gonna be found in the research paper.
If you just feed it to the ai, you have to kind of tell, okay, what should this look like? What is usable? So there are many different types of context engineering that we're using in NISA, but that is sort of the core. And I think it's really exciting also because you know, some of these core practices, as we talk to teaching lab experts, we sort of land on a set of core practices of like, okay, this should be core to NISA and all teachers, all the advice we're giving, all the resources we're creating should be drawing from this.
But of course, different schools, different districts, different coaches, they also have different other resources, right? Sure. Curricula, some are using the Danielson framework. Some have different contexts in their schools, and maybe they have their own
[00:37:09] Alex Sarlin: portrait of a graduate they can put in there. They
[00:37:11] Louisa Rosenheck: have, right, they have their own priorities for learning goals.
They have their own instructional practice guides. And so another great thing about AI is that you can swap those resources in and out. So that's a system that we're building within NISA. Where NISA can have different parts of its brain for different users or for different contexts. That's really powerful.
And that can be, you know, at this like district level, but it can also be at the level of a coach, like kind of more in line with Enlighten AI. You can train it on, you know, what a, a certain. Grading style that you, that you want everyone to calibrate to, and you can give it, you know, this is the way I like to coach.
These are the things that are most important to me. And, you know, for even at, at that level. So that, again, it goes back to what I, what I said initially about the goal of NISA. It's not just a, a tool to make things more efficient or to give you the best answer. It's really meant to feel like things are coming from your coach and to, to strengthen that connection between the coach and the teacher.
Because we know from research that that is one of the things that. Uh, really forms a powerful coaching relationship and, and makes the coaching more effective.
[00:38:20] Alex Sarlin: The concept of context engineering. I've never heard that phrase. Maybe I'm just behind on it. That is, I think, an incredibly interesting way to look at how this all works.
So, you know, we think about different types of, of instruction or prompt engineering or sort of setting an AI up to do what you want it to do. But context is key in education and context engineering. Talking about what is going on in the environment feels like core to your project, but it feels like.
Core to what everything that's happening at Teaching Lab.
[00:38:45] Louisa Rosenheck: I'll say context engineering is a term that's used quite a bit more and more now in ai, but not as much in in ed tech. I think it's not enough a part of the conversation in Ed Tech and on NISA along with, uh, other teams in teaching Lab Studio.
We're planning to write a white paper about this, so I'll share that when it's ready and that'll be a resource to, you know, try to help other Ed techs think about this. And of course, you know, improve. The context of their products too.
[00:39:12] Alex Sarlin: It's huge. I mean, it's also relevant to what we were talking about with Coteach about, you know, the context of a classroom.
You know, a teacher has the same classroom or the same set of students usually for quite a while. Knowing that set of students, you know, knowing enough about them, knowing their, what they need is context that can tailor instruction. It's, that's true for sort of any differentiation based product. The concept of swapping out parts of your brain really sticks with me, and it's a really, it's a funny metaphor.
It brings to mind another funny metaphor that I think about with ai. This is gonna lead to a question here, which is that I sometimes flash back, if you remember the short circuit movies, I think from the late eighties, I don't know if anybody's that world, these movies about these robots that come alive.
And there's a moment where like Johnny Five, the robot is like faced with like people who are trying to fight him and he picks up like a kung fu manual. And because he's a robot just looks through the whole kung fu manual and suddenly knows kung fu and can fight them off. And I always think of AI and I'm like.
That suddenly is completely real. That idea that you can take a, like a universal brain and say, no, I want it to be an illustrative math expert today. Now I want it to be an incredibly calibrated grader. Now I want it to be an instructional coach, and if you give it the right material, the right data, and the right instructions, it can.
Instantly be helpful and, and even outperform humans in some cases. It's just such a wacky time we live in where that's suddenly true and I think we're all figuring out what to do with it. So here's my question. You three are amazing fellows. There are a bunch of incredible projects happening at Teaching Lab Studio all around you, and I know it's such a collaborative and environment.
I'd love to hear each of you talk a little bit about other things that are happening at Teaching Lab Studio, just to help our listeners sort of illustrate the breadth of different types of projects that are happening. 'cause there are some really interesting ones. Out there that compliment what you are all doing.
You got them. Can I actually start with you on this one? Tell us about what else is happening at Teaching Lab Studio that inspires you to collaborate or just you're excited about?
[00:41:05] Gautam Thapar: Yeah, there's a few that came to mind, so I'll try to pick the one that's most adjacent to Enlighten AI, but if nobody talks about Math Lee or some of the other great projects come back to me.
But one of the really interesting projects that actually was. Part of what attracted me to the teaching lab in the first place is called the Writing Pathway, and Phil and Sherry, who lead the, the, the Pathway and yc, who, who's been doing a lot of the development, basically constructed a version of what we've been talking about for writing resources to really build writing skills.
So Phil and Sherry are both writing experts and. Really valuable data sources. Like what would an expert actually put in front of kids to build their writing capacity? And they basically codified a curriculum, a scope and sequence that cuts across every single curriculum state. You name it, it's just, it is how to do writing.
The sequence of things that a student needs to know and be able to do in order to to write. And then the ability to take the context of the classroom and create a writing resource on any specific skill. And so, you know, we've been learning a ton from them in terms of how do you take the data when a student has written something and figure out where to sort of map that student to, in terms of the resource, the right level to support them.
Is it sentence level? Is it paragraph level? Is it structure? And so I think they've just done some incredible research They have. You know, it's a proven efficacy and just really incredible work.
[00:42:29] Alex Sarlin: Yeah. And when you, you focus on a particular area, like writing, bring everything together, create a meaningful corpus and sort of curriculum around it, it benefits the entire field, which is just so exciting.
Riz, how about you? What other projects within the Teaching Lab Studio inspire you?
[00:42:45] Riz Malik: Yeah, I might mention to the other kinda math focus tools at the moment. So one of those is, is Math Lee, which I really love. So, so they've taken some of the science of learning research around, worked examples and they're kind of building out a really, really cool kinda innovative student practice and learning platform.
They kinda is grounded in this work. Example research, they're a bit earlier on. They're doing lots of kinda co-design and rapid testing cycles at the moment. They're testing with, with lots of kids I think in your city at the moment. And they're doing really, really interesting work there. And then the other one that I might mention as well is, is pods C, which is also taken a very kinda research driven approach this time about the, the kind of the retrieval effect, another kind of findings from cognitive science.
And I think about how do we kind of do student practice in a way that becomes sticky over time. And so they're, they're doing some really awesome work as well. And I think, you know, the reason I mentioned those two as well is we are thinking a lot at the moment about what it might look like to share data between tools and between products.
And so kinda one of the exciting visions we've all got is like, what would it look like if kinda math Lee can talk to Posy and Coteach, can talk to Posy and Math Lee and there's a kind of this kinda movement of data around so it becomes more coherent. And so yeah, both of those things are really, really exciting at the moment.
[00:43:57] Alex Sarlin: Yeah, as a wise person. One said, you can't even buy coherence right now, so we've gotta find ways to build it. Louisa, what other projects within the Teaching Lab Studio inspire you?
[00:44:07] Louisa Rosenheck: I was gonna talk about math as well, because the project I mentioned about engagement, motivation research is collaborating with math.
So one of the things that we're, we're doing is seeing, okay, what are the things that, the approaches that we're, we are researching and how could they be used in that context? And Riz mentioned briefly. Both of us are doing a lot of co-design on that project, so with middle school students and with teachers, and we're really learning a lot from each other and, you know, sharing information and ultimately the product and features that we create.
So that's a lot of fun and a couple of other tools that I'll mention. A couple of these are, I think, have, one thing they have in common is student agency, so students really being able to make choices and contribute, share their own voice. So one of them is Study Buds. I know that you had Adam Franklin on the podcast, so you're familiar with that one.
Just to recap, at the core of that, is the student kind of teaching. An AI character, an AI agent, helping that character understand the content. So it kind of gives, it puts the student in different kind of role. So that's great. And then another one is called Super Structures. Super Structures is playful and fun.
So you know, something that I really believe in it is social. Everyone is contributing to their responses, to a prompt in different kinds of graphical organizers, different structures. And so it's really a way that everyone in the class. Can contribute their ideas and then build off of other students' ideas, which is, you know, a type of social learning and, and engagement that I think is so important.
So that's an exciting one, and that, that project also, the way it came about is kind of unique to Teaching Lab because it was a bunch of different fellows that were working on individual projects and then had this idea and came together and they were able to make it happen. And it, it has launched recently, so it's out there.
Everyone can use that one too.
[00:45:58] Alex Sarlin: Yeah, that's an exciting one and I think it shares a little bit of DNA. So you also have a background. You were at Kahoot supporting them with their learning engineering and sort of figuring out how, how to inject more learning into their world. And I feel like Superstructures has some of the same DNA as a Kahoot in that it's like whole class and that's social flavor.
[00:46:14] Sarah Johnson: Mm-hmm. Social
[00:46:15] Alex Sarlin: play, very engaging. One thing that strikes me as I hear you each talk about different, some of the different projects is that each of you are working on projects that are primarily. Teacher facing, which makes a lot of sense and is also, I think, you know, shares a lot of DNA with, with relayed graduate school and with teaching lab.
But some of these other projects are student facing. I think we're in this interesting moment right now. We've done the, these sort of market maps of the AI space and we're in this funny moment where I think people building for students. In school, people building for students outta school and people building for educators.
And, and schools are really, actually think all thinking pretty differently about the AI space. They, because people have different incentives, they want different things, they have different constraints. They have, there are different laws. There's all sorts of different, you know, the human in the loop is incredibly important in, in, in anything related to students.
In school, but they're not out of school. I'd just love to open that up for a moment because I think it's a, just an interesting aspect of the AI space in this moment. And because Teaching Lab Studio sort of thinks about it through multiple lenses, I'd love to hear each of you just talk about it a bit.
And just to be clear, the question is really like, we're in this world where people who are building for out of school, students are building tutors and study tools and things that maybe, you know, take a picture and it'll give you an answer. People building for students in school have to think about all of these.
They have to be very careful. They're trying to empower students, like you just said. They're trying to make it engaging and fun and interesting, but also keep the teacher, you know, at the heart of the experience. And then people building for teachers are solve, trying to solve problems for teachers, like all of you, like really serious problems.
Those are all really different and I'm curious how that could come together to create a sort of cohesive ecosystem. Because I know it's a little bit of a high level question, but I'm curious what that it brings up for you.
[00:47:54] Riz Malik: Yeah. The first thing for me is I'm imagining that the, the, the teacher is like the orchestrator or the conductor in the classroom.
Okay. If you take that analogy. And so the teacher is the one who's ma who's gonna, in the driving seat, that they're kinda confirming those decisions. They're checking things over. Anything that's going to, the student is gonna being approved maybe by the teacher, and they've kinda got some oversight of what's going on.
I think however, this kinda pans out, the teacher in my view, gonna need to be at the center of everything and they really play a really, a really central role in all of this. I think. I'm personally happy that kind of with our project right now, we're just trying to think about teachers. There's so much more complexity I think that comes up when you're trying to serve, serve students.
And I know that Gautam's been facing a lot of that in his work kind of recently in some of the, the additional kind of compliance and challenges that come there. Maybe I can share Hannah as Gautam. You can share some of that. Sure.
[00:48:40] Gautam Thapar: Yeah. Uh, well, Riz was saying really resonated and I, I think what stood out in your question, Alex was around this like human in the loop and I think.
One of the things that I love working in the teaching lab is we really think a lot about what it really means to be in the loop. And so one version of being in the loop is like you pass your student writing over to the AI and it passes you something back and you're checking it, you're in the loop. But that's a really unexciting and uninspired way to be in the loop.
And so I think one of the concepts that cuts across a few of the products is this idea of how the teacher is in the loop. And I think there's sort of like a really important beginning where the teacher is infusing context. For the AI in some way, shape, or form. Right? And one of the most valuable data sources is what the teacher actually thinks.
The right thing. The the right move is for the classroom and how the teacher's in the loop at the end to sort of review and, and assure and give feedback and make sure that the next time it's even better. And I think that's a much more interesting way to keep them in the loop because there's a lot of rote work in the middle, combing through the PDF and R'S case to find ims answer to a particular question as opposed to asking it.
You know, just that change makes it so much more accessible. And so I think that's a huge piece of the puzzle. 'cause like Rose said, I think the teacher is ultimately has the most knowledge, the most information, the ability to make the best decisions. And if you've ever subbed a class, even one where you know, the students like I have, like, you know, that the teacher actually does, um, just have this vast knowledge and getting that into the model in some way, shape, or form to assist.
Is just so, so crucial.
[00:50:13] Alex Sarlin: One thing that excites me a lot about your, your model, and it comes up in that answer, is that teachers are are educators. Educators are educators. They wanna educate and they wanna educate their students. But the paradigm of having them educate the AI of how to do better, of how to go deeper, of how to, you know, add more value to the classroom, is a much more powerful metaphor for how a teacher can work with an AI than.
As you say, being a a checker, being the sort of, oh, the AI did 90 per, 95% of the work and I just have to read it over and hit submit, which is how many of the tools work. I think that's a really powerful metaphor. The idea of put putting the teacher in much more central, much more in the driver's seat, but also allowing them to, to educate.
Educate the AI to become a better assistant, to become a better arbiter of content, whatever the AI is actually doing. That's really, really interesting. I know we we're almost at time here, but Louis, I wanna bring you in on this as well. You think a lot about student facing ai. What do you see when you look at the landscape right now and look at how different tools look when they're designed for students or for teachers?
[00:51:16] Louisa Rosenheck: Yeah. I mean, I think like the others have said, the most powerful learning would come from a system that's all interconnected. Right. And on the, on the Ting and Thrive project, this is something that we're thinking about. For example, we, one of our prototypes we're testing out is about family engagement.
So, you know, how can we let parents know what's going on in school and give them little playful activities that they can do with their, with their kids to engage them. That's gonna be the most meaningful, the most authentic. If it's connected to what they're doing in the classroom. Let's say they're using math, lead math could inform this system and, and it could know what are they struggling with, what are they having success on.
So for sure, you know, sharing information would make a more coherent and meaningful experience for everyone. And with, you know, with these examples in mind, just to kind of pull back to the 10,000 foot view, I think the way the ed tech field. It has developed, it really disincentivizes any of this, you know, tools are so fragmented.
Every company is doing its own thing and I don't blame anyone for that. That is of course, that's how it works, but there's no incentive to connect with other tools. It's of course very hard to share data, rightly so. So that's a problem that would really need to be solved collaboratively. It's, you know, the experiences are different.
You can only connect things within a suite of tools and even then, a lot of times it's, it's sort of, the products feel very separate. So I think the way the EdTech field and, and you know, EdTech companies have been set up makes it even harder for these connections to happen. So I think it's another conversation.
You know, what could be done about that? But just to say that there's a lot of really meaningful approaches that could be taken and each company is sort of focusing on, like you said, they're focusing on the student experience or they're focusing on the teacher experience and like they get really good at that thing and they have to, in order to make their product really good and, and get traction and be sustainable, but then where's that holistic view?
I think that's something that's very, very much missing in EdTech and very hard to create right now.
[00:53:22] Alex Sarlin: Yeah, and people are hungry for it. I think you know that that concept that Sarah said of an AI centric school where you know you have multiple tools working in coherence is one aspect of it. You might also say that these sort of tool suites are an attempt to try to do that, and they become popular because they're an attempt to try to do that, even though.
There's no real data sharing happening under the hood. I'm not sure I would call it coherent teaching, but the idea of having many different tools, all working with the same teacher in the same classroom, at least gives the illusion of that, but we're not there yet. I think Teaching Lab Studio is really at the forefront of that type of collaborative approach.
This has been such a fascinating conversation. I'm so glad to have you all. On the podcast, if anybody who's listening to this who is not already on the Teaching Lab studio site, who's not already checking out all of these products that these amazing fellows are building, and the ones that their their fellow fellows are building should definitely be there.
Thank you so much to Dr. Sarah Johnson, who's the CEO and President of Teaching Lab Studio to Riz Malik, who leads Coteach and is a fellow at Teaching Lab Studio to Gautam Thapar, who is a CEO of Enlighten AI and Teaching Lab Studio. And. Louisa Rosenheck, who co-leads NISA and Tangle and Thrive within Teaching Lab Studio.
So great to have you all here on EdTech Insiders. Thank you so much, Alex.
[00:54:37] Louisa Rosenheck: Thanks for the great conversation.
[00:54:38] Alex Sarlin: Thanks for having us. Thanks for listening to this episode of EdTech Insiders. If you like the podcast, remember to rate it and share it with others in the EdTech community. For those who want even more, EdTech Insider, subscribe to the Free EdTech Insiders Newsletter on substack.
