Week in Edtech 8/19/26: ChatGPT for Teens, $265M Reach Capital Fund, The EdTech Backlash, AI’s Competitive Shakeup, and More! Feat. Sabine Wallis, NWEA & Andy Rotherham and Marisa Mission, Bellwether

Join hosts Ben Kornell and Alex Sarlin as they break down the biggest stories shaping AI, Edtech investment, the growing tech backlash, and the future of assessment.

✨ Episode Highlights:
[00:05:47] ChatGPT for Teens introduces new safety and learning features
[00:11:29] Generalized AI models raise new challenges for Edtech companies
[00:13:47] OpenAI partners with CodeAI on AI learning and youth safety
[00:17:07] Reach Capital launches a new $265M fund focused on early-stage companies
[00:23:32] B2B vs. B2C: New dynamics reshape Edtech investment
[00:25:12] Hugh Grant and Closed Screens Open Minds fuel the Edtech backlash
[00:27:46] Why Edtech needs to put education first and technology second

Plus, special guests:
[00:32:28] Sabine Wallis, General Manager of NWEA, on assessment, learning science, and using data to drive instruction. 

[00:51:23] Andy Rotherham and Marisa Mission, of Bellwether, on AI in education, EdTech quality, market sustainability, and designing for productive struggle.

😎 Stay updated with Edtech Insiders! 

Follow us on our podcast, newsletter & LinkedIn here.

🎉 Presenting Sponsor/s:

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.

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.

Tuck Advisors was founded by entrepreneurs who built and sold their own companies, frustrated by other m and a firms, they created the one they wished they could have hired but couldn't find. One who understands what matters to founders and whose North Star KPI is the percentage of deals closed. If you're thinking of selling your ed tech company or buying one contact Tuck advisors now.

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 RFP stage. That’s how top edtech teams stay ahead.

Transcript

Copy Transcript Copied

[00:00:00] Alex Sarlin: Innovation in pre-K to grade learning is powered by exceptional people. For over 15 years, Edtech companies of all sizes and stages have trusted Hire Education to find the talent that drives impact. When specific skills and experiences are mission-critical, Hire Education is a partner that delivers.

Offering permanent, fractional, and executive recruitment, Hire Education knows the go-to-market talent you need. Learn more at hireedu.com. That's H-I-R-E-E-D-U.com.

I mean, if you think about what the experience of being in ChatGPT for teens is, it's still back and forth text. Even if they're gonna create a visualization, it's still pure back and forth text.

Yeah, I guess you could do some voice, but it's not immersive. There are some really interesting Edtech tools coming out where the actual user experience is much more engaging and connected, and it's not just like a back and forth text message app. And yes, ChatGPT could move into that direction, but I think there's still an area to differentiate, and I've seen a lot of really interesting Edtech companies who are thinking about both the pedagogy but also the actual, like, what are you doing when you're engaging, when you're connecting with my AI companion or with my AI flashcard app or with my AI study tool or with my AI tutor?

What does it actually look and feel like? And I think there's still, at least for now, room to innovate there.

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. 

[00:01:42] Ben Kornell: 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 backchannel insights from Alex and Ben. Hope you enjoy today's pod.

Edtech Insider listeners, we are here. It's Week in Edtech a day early. We're so glad to bring you the latest and greatest in all Edtech news. I'm Ben Kornell, along with Alex Sarlin, and we are Edtech Insider co-founders. Few shout-outs before I hand it off to you, Alex. We've got a Thursday happy hour in San Francisco coming up.

If you're listening to it, it's probably tonight. And then also we have our Edtech Insider WhatsApp channel. If you subscribe to our newsletter, you automatically get to join our Edtech Insider WhatsApp channel. It's where all that conversation and action is happening. And then last but not least, we have two big, big announcements.

September, we have a happy hour in Los Angeles co-hosted by Jacob Kantor, DODO, and a group of other Edtech organizations to be shared soon. And October first and second, we're at the Teach for America summit, and we will be on stage interviewing Lin-Manuel Miranda. Few of you may know that Lin was actually a teacher before he became this famous intellectual in American history.

And he has deep commitment to arts education, deep commitment to education for Puerto Rico. He also has a lot of interest in history and Hamilton, EduHam. So we're gonna be diving into all of that live on stage in Caesar's Palace. I believe that's October second. So we're gonna be bringing that to you. So much exciting stuff.

Before we go into the news, what else is coming up on the pod, Alex? 

[00:03:41] Alex Sarlin: Yeah. I mean, next week we have this fantastic interview with Tony Wan and Betsy Corcoran of EdSurge, the founding team of EdSurge, talking about the EdSurge story, everything that happened to grow EdSurge, and all the memories there and all the things that happened just to become like the defining Edtech publication.

Really, really interesting. We're talking to Sabine Wallis from NWEA, now part of Houghton Mifflin. NWEA MAP is still like the test of record throughout many, many K-12 schools in the country, and we're gonna talk about what that means and how it's evolving in this current age. We're talking to Carl Madi from StepFul very soon.

We're also talking to Mario Vasilescu from Readocracy. They have a new paper out all about ethical AI and how we can understand information in the age of AI. We're also talking to Bellwether. I mean, it's nonstop interesting people on the podcast as always. But yeah, it's gonna be fun to be out in TFA as well.

And it looks like I may be going to an education conference in September up in DC. I'm looking forward to reporting on that one as well. So much cool stuff happening in this space. 

[00:04:43] Ben Kornell: The other thing I would say is we're having university presidents coming to talk on August twenty-eighth. We've got Oklahoma University President Joe Harroz.

We've also got Lisa, the new president at Southern New Hampshire University, and our good friend David, the CEO of Masterclass, joining us to kick off a special series where we're gonna talk about the transformation of higher education. Matthew Rascoff from Stanford is teaming up with us to bring you that special series.

And we hope you'll join us there for that webinar. So keep tuned to the Edtech Insider WhatsApp channel, our LinkedIn. That's where you're gonna find the latest and greatest on what we're up to So now we're gonna jump in, and I will say, what a week it's been. We've got Reach Capital launching their new fund.

We've got ChatGPT for Teens, and we've got the latest in the tech lash with Hugh Grant, of all people, coming. He used to be our star-crossed lover, and now here he is coming in, slamming all Edtech. Where do you wanna start, Alex? 

[00:05:47] Alex Sarlin: Let's start with the ChatGPT for Teens announcement because this is really breaking news.

It just was announced today, the day we're recording this, and it's a really interesting one. It's interesting because it is very much a combination of safety factors. You can imagine exactly the parental controls and the type of guardrails that would be built into something like a ChatGPT for Teens in terms of mental health, in terms of early alerting systems, in terms of parental auditability.

I mean, there's a lot in there, but there's also a pretty significant learning component, which I think is relevant for all of us in the Edtech space, just to see how they're thinking about their own learning responsibility as pretty much the most used AI program in the world. Do you wanna talk through some of the learning pieces?

Or Ben, what was your reaction to when you saw this news come out? 

[00:06:33] Ben Kornell: Well, I think first, the reality is most teenagers are already using the full-on ChatGPT. So I would say my first reaction to this kind of stuff is I wear my Common Sense Media hat, where I worked for a couple of years, and there's the, like, what's published, and then there's reality.

And the reality is that my teenage son, who's 14, probably isn't gonna be happy with something that's super guardrailed. He's gonna want the real thing, and he's gonna figure out how to work around it. And so the question is, like, how young is this thing gonna go? Are we gonna have 10-year-olds or eight-year-olds or six-year-olds using this?

And unfortunately, I think that's the reality of when it hits the road. And so when it says, "Built for learning, backed by protections," I think that actually raises some more concerns. I also think that the theme that we've talked about over and over is purpose-built AI tools for kids outperform those that are generalized.

This does not feel like purpose-built. It feels like harnessed, and I worry about how strong those harnesses are going to be for the functions that we're talking about. But I think anybody who's working in schools is very unlikely to jump into the deep end on that. I think for parents, they've probably already jumped into the deep end, and this will be just like YouTube for kids- Yes

which has very, very small penetration of a few high-oversight parents, but the rest are on the generalized models. That's my cynical take. 

[00:08:11] Alex Sarlin: But that's fair. I mean, I think some of the things that are interesting about it to me, if you just look at it from an announcement perspective, how they're framing the education pieces of this.

So ChatGPT, as we know, has created a study mode a while ago that's trying to make it more Socratic, trying to base it more in learning science. And for the most part, the main thing it's trying to do is try to steer students away from asking AI to literally do the work for them. And some of the pieces of this announcement are saying how they're going to basically recognize, they've tried to train the ChatGPT for teens to recognize when a teenager is asking them to actually do the work for them and redirect them towards study mode.

One piece of it, and that the proof is in the pudding there, but it's a noble attempt at least, and something they've heard from lots of different people about the academic integrity concerns about ChatGPT, which are very real. One thing that's interesting about it, we announced a couple of weeks ago how Google had taken their guided learning mode, and in the new Gemini classroom pieces have created the ability for teachers to basically turn on guided learning, to basically make it the default, almost force students into guided learning.

That's partially as just an administrative tactic, but also I think a recognition of exactly what you're saying, that it may not be something that teenagers are actively opting into. They're not necessarily choosing the learning mode of Gemini or in this case of ChatGPT. So what they're doing is they're actually creating a concept called study hours, where either parents, probably mostly parents or teenagers themselves can say, "During these times, you can only use study mode."

So if you're on ChatGPT doing your homework between 4:00 and 6:00 PM, you're in study mode. You don't have a choice. And that, I think, to your point, it's probably for high oversight parents. This is something, it's like a parental control that is baked in. There may be a little bit of window dressing or CYA thinking there where they could say, "Hey, look, if a teenager does something really strange, well, the parent didn't turn on study mode," or there's a little bit of a passing the buck piece of it.

But I think it's also a positive move that these tools are recognizing that unprompted, that without any guidance or without any oversight or without any sort of parental involvement, young people really are going to the least common denominator and asking for homework help and answers from a lot of these systems.

And they're starting to at least provide the opportunity for parents or teachers to limit the types of modes. It's not, to your point, purpose-built for education, although they have introduced quizzes, they have introduced visualizations, they have some learning features built on top of, like, as you say, a harness on top of traditional GPT.

It's not a fully separate product. But I do think it's interesting that they're at least moving in the direction of trying to make it less of a complete free-for-all where young people just open ChatGPT, can ask anything, can do anything, and it just sort of directly interferes with any kind of educational process.

So I think it's a step in the right direction. I think probably the main impetus between ChatGPT for teens was the safety piece and the PR piece, and not the education piece. But the OpenAI education team is a real team. They do think a lot. They're very smart people there, and they do think a lot of education, and I think they're trying to inject an educational component into the ChatGPT for teens, which hopefully they will continue to evolve on and to test.

I mean, I'm curious about that. Are they going to actually research whether ChatGPT for teens or study mode has the effect that they really want? 

[00:11:29] Ben Kornell: Yeah, it does seem like you can't dip your toes in this space, you really need to be in. And this pairs well with the ChatGPT for teachers in that you could imagine a B2B product and a B2C product, and those are kind of twinned that way.

And the cognitive friction is dialed up in the school context, and it's only dialed up when you're in the study mode on the GPT side. It's interesting that the on-label use of Claude is still eighteen and over as a policy. So I wonder whether they'll come out with a consumer-facing teen mode as well, just as Claude for teachers was kind of in response to what OpenAI did with teachers and the AFT.

You know, in general, one challenge for the Edtech industry is that this stuff starts competing psychologically with Edtech tools. And so if you spent all this time and energy creating a kid-safe AI chatbot and now this is coming out, are people going to switch from your product to this product or vice versa?

I don't know. They may just see a lesser value of your new product given that they're like, "Well, I could always just do the ChatGPT for teens." And this is one way in which YouTube really suppressed other video services that were specific to kids popping up, is they'd be like, "Oh, well, if that's what you want, we've already got it here, and we've got the world's largest library of YouTube for kids."

And so in that sense, it can actually warp the competitive environment. And I still come back to the only defensibility in the Edtech space against these generalized models is really the pedagogic precision, the expertise around learning friction and cognitive lifting, 'cause it's really hard to have like a-- even these, I'm seeing a lot of like kid companion AI that bots They're just so hard to compete against generalized models or against the latest character AI.

If your business becomes a big thing, then all of these bigger players who have way more distribution, market share, and brand, all they have to do is just launch a teens mode and you're crushed. So I think it's a tough space to be an entrepreneur in right now. 

[00:13:47] Alex Sarlin: It is. One more piece of this news that I think is worth calling out, especially for our audience, is that as part of this announcement, they also are announcing a signature partnership with CodeAI, which involves a variety of different mechanisms worth looking up.

But CodeAI is formerly Code.org. They have the Hour of AI. They're doing something called the Builders Challenge, in which high school students have the opportunity to create with AI and have mentorship with AI team members. There's some AI training, and they have this concept, to your point, about a joint advisory council that will, in quotes here, "Bring together experts and leaders on child development, youth public policy, and learning science," so that there's a little bit of a oversight committee between OpenAI and CodeAI, which is a nonprofit, but very much in that deeply embedded in all of the tech company space and has been for many years, thinking together about what responsible AI innovation looks like.

I think to your point about market distortion, I mean, this has come up many times in this podcast, and it is very tricky for a lot of reasons, and we can revisit this as well and through the lens of all the AI backlash that is happening. But I think that one of the things that's really confusing right now is that Edtech companies are under pressure from both sides.

They have these foundation frontier model providers who are creating platforms, and increasingly you can see why, right? They're creating platforms. This ChatGPT for Teens, it is a slice of the ChatGPT ecosystem with harnesses on it, with a sort of educational, at least veneer. Hopefully it's deeper than a veneer.

And the question is, as you say, if you are a family who does want to use AI or does want your student to have access to AI for their learning, do you just turn on study hours for ChatGPT for Teens, or are you looking for deeper solutions that may be built on open source models, or they may be built with standards integration, they may be built with really deep learning science thinking inside them, like many of the companies we talk to on this platform.

It's hard to compete upward, but what we saw with the Claude announcement is there may be space to actually work together, to plug in, to actually use the distribution of these giant... You could imagine ChatGPT for Teens having plug-ins that are specifically education and teenage oriented. They may be for mental health, and they may be for education, and those could be some of the top Edtech companies.

So it's tricky, but they're also under pressure from the bottom because you have people using vibe coding apps and creating all sorts of new Edtech tools. So it is not an easy time to be an Edtech company. We advise entrepreneurs, we support a lot of different people. We invest sometimes, but like, it is not an easy time.

That said, I do think there still are more areas of differentiation potentially even than pure pedagogy. I mean, I think there are experience differentiation aspects. I mean, if you think about what the experience of being in ChatGPT for Teens is, it's still back and forth text. Even if they're gonna create a visualization, it's still pure back and forth text.

I guess you could do some voice, but it's not immersive. There are some really interesting Edtech tools coming out where the actual user experience is much more engaging and connected, and it's not just like a back and forth text message app. And yes, ChatGPT could move into that direction, but I think there's still an area to differentiate, and I've seen a lot of really interesting Edtech companies who are thinking about both the pedagogy but also the actual, like, what are you doing when you're engaging, when you're connecting with my AI companion or with my AI flashcard app, or with my AI study tool, or with my AI tutor, what does it actually look and feel like?

And I think there's still, at least for now, room to innovate there. 

[00:17:07] Ben Kornell: Yeah. I think this leads actually well to our conversation about the investing space We've been talking about an Edtech winter for quite some time. Funding has gone down and we never recovered back to COVID levels, but even pre-COVID levels, we haven't really seen investment jump back up.

If anything, I'd say Europe is the bright shining star in Edtech investment circles in terms of growth. And of course, we've had some pretty big scandals including, like, Byju's, and we've had Paper and its crash. So against that background, it was exciting to see Reach Capital announce their Fund V. And we had Jomayra on the podcast, so everyone check that out.

I'm sure you've seen us promoting on different channels. But part of our excitement is because Reach is really a success story in terms of Fund I through IV, and also it's a bright light on where Edtech investing is heading, which I think is earlier stage, pre-seed, seed, and A. There's still An opportunity to accelerate high impact programs and also make money doing that.

But I think this round is actually a pretty big moment for the space in that GSV has kind of receded. Owl is still sitting on a really big fund and not clear when they're going to raise again and what, what that will look like. And Reach is kind of emergent as the premier Edtech fund in the space.

There was a lot that we asked Jomayra. I'm curious, as you've had time to process it, like, what do you think are the big takeaways for the space with this new announcement? 

[00:18:48] Alex Sarlin: The big one that she talked about in that episode, and I'd recommend everybody check it out, that just came out just a few days ago, is that they're actually going even earlier stage than regular, th- than they had been in the past, which I think was...

You know, you look at an Owl, which much of its distribution has gone to the success stories of previous rounds, and it's been to sort of double down and follow on rounds. But Reach is still very excited. I mean, this fund will write checks of $1 to $10 million spanning pre-seed through Series A. That pre-seed is much earlier than they've done in the past, into roughly 50 companies over the next three years, and none of whom have received funds yet.

So we're talking about 50 early stage Edtech companies that Reach is going to be funding over the next three years. That definitely, as you say, makes them sort of the most active early stage Edtech investor out there. And for many of the people who listen to this podcast, that's, they're scribbling that down right now, or they're looking at the announcements right now because they're like, "Oh, wow, 50 pre-seed to Series A companies.

That's a lot of people." Now, we should also say that Reach has expanded its thesis. This used to be only Edtech specific, and now it's education, health, and work, which, so you can imagine that of those 50 companies, that it's gonna be a distribution. Some of the limited partners are worth talking about.

They, they said they raised this round in six months. It's a $265 million round from places like the Lego Foundation, College Board, LA Fire and Police pensions. But I think that it's really, they brought on some new folks. But I think what's really interesting about this from what we heard from Jomayra and what you sort of see out as they're announcing the round is that, you know, they've been successful in many ways.

A lot of the companies that have really sort of burst into the space as Edtech success stories have come through the Reach ecosystem, and they're really looking to continue to be sort of people who support the early stage companies as they go in and not just do a strategic... You know, we talked for a long time, Ben, about like people doing a wait and see in the AI era.

This idea of like, we don't know what's gonna happen. We don't know who will survive the frontier models. We don't know who, what differentiation and looks like from a distribution standpoint or proprietary data standpoint in AI, and I think this is a signal that Reach is not waiting anymore, right?

They're like, "Over the next three years, we think we're gonna find 50 companies, early stage companies that are VC worthy." That's big news 

[00:21:06] Ben Kornell: Yeah. And you know, the reality is the formula at Reach Capital only partly has to do with Edtech as an industry Their theory is really find really great founders really early that are, like, mission obsessed.

And I think the application of regulation and unique market dynamics creates defensibility is applicable in the Edtech space, but it's also applicable in the healthcare space. So if they didn't have that laser sharp focus on awesome entrepreneurs that they're backing from the beginning, I don't think their thesis actually works or plays out.

So if I were to factor out like what their success is attributable to and where they're playing on this one, it's a significant-- A preponderance of the success is around founder selection and around like core business model defensibility, and then it plays out in a lot of different ways in Edtech and in a lot of different ways in healthcare.

You'll note that they have some bets, uh, Pathfinder as a ESA play, and they're in Outschool as like, you know, direct to home. And then they've had a number of B2B K-12 solutions. The other thing that I'm thinking about a lot is do we actually have room for series C and series D companies in Edtech anymore?

Good question. And you know, to be fair, and I'm not sure what Reach's, um, stake is in Stepful 'cause they were one of the more recent ones that raised at a higher valuation, but they exited at like prime time from paper. They, from what I understand, they also, you know, have been backing some companies that have pivoted from an Edtech narrow to like broader tech over time.

So I just don't know if we can support these like later stage rounds anymore, and we're definitely not seeing billion-dollar exits anymore. So that's an interesting point. 

[00:23:12] Alex Sarlin: That's true, but Jomayra also mentioned that they have had six exits in the last six months, not billion-dollar exits necessarily. Not billion, yeah.

Yeah, but there are M&A happening. We saw ChatGPT Zero get acquired, which is one of their portfolio companies. So there are outcomes, there are meaningful outcomes, even if they're not, you know, billion-dollar and they're not IPOs. But keep going, please. 

[00:23:32] Ben Kornell: The other thing that we talked about with Jomayra was B2B versus B2C, and I think for the last decade it's been all about B2B because ARR was kind of the coin of the realm, and when you're selling to schools or universities or workforce, that reliable ARR is backable from an investor standpoint.

But Jomayra kind of shined a new light on B2C as this like incredibly promising category in education and healthcare because we're seeing things ramp so incredibly fast in terms of consumer adoption. And meanwhile, on the B2B side, we're seeing like cost and price compression So that's gonna be an interesting story as this plays out.

For our listeners, just, you know, the dynamics of a pre-seed and seed fund are the outcomes play out over a decade, and Reach has been around for eleven years now, so they've seen some really great outcomes from their earlier investment portfolios. But we won't-- The story won't be told on this round of investing for another five to seven years when we really see where things head and how AI ultimately changes our space.

With that, I think the other transition is how is AI and technology changing our space? 

[00:24:46] Alex Sarlin: I have a segue for you. I mean, Reach has been really clever in following the trends, and one of their recent investments was a company called Doorman in twenty twenty-five, which is all about blocking, quote, "high dopamine distractions."

It's basically a cell phone blocking system. It's a school infrastructure service to block distracting cell phones, and I think talk about following, you know, where the puck is going. I think, you know, we are in a very strange moment. I think you were gonna segue to the backlash, right? 

[00:25:11] Ben Kornell: Yeah. 

[00:25:12] Alex Sarlin: Yeah. So the backlash is real.

We've seen some boldface names over the last few years, like, of course, Jonathan Haidt and Jean Twenge, who are researchers, Wayne Holmes, who is at a, a very UCL AI researcher who has a, a lot of warnings about the, the negative effects of AI, and Jared Cooney Horvath, who's gotten-- made a name for himself with his book "Digital Delusion," and he's testified in Congress.

He's a neuroscientist who's basically trying to pin all of the declines in schooling on technology. You can decide for yourself how you feel about his cases. But there's a new face in this world, which is actor Hugh Grant from all sorts of amazing movies, and also actress Sophie Wilkerson, who's a British actress as well, and the two of them have started an organization called Closed Screens, Open Minds that is very actively saying that the entire Edtech industry is out to collect data, hurt kids, and just completely-- It's an anti-Edtech organization.

[00:26:10] Ben Kornell: They're basically saying intentions are explicitly bad here. Everything with education and technology is wrong. It does point to the full extent of backlash, which is tech abstinence is, I think, what they're advocating up through, you know, elementary school essentially. You know, I think they're leaving no room for any value add of technology that faces students or young kids.

[00:26:42] Alex Sarlin: Yes. It's becoming a very serious chorus of folks who are trying to warn the world about, quote, "Edtech's dark side." I have lots of opinions about this, as you can imagine. I'm sure you do too. I'm sure many of our listeners do too. I mean, one thing we should just acknowledge straight off the bat is that there have been declines in test scores.

There have been declines in IQ. There has been a lot of negative effects attributed to cell phones and to social media, and there's been depression and anxiety, and, like, there have been lots of negative effects. Arguably, screen time in school has gone up a lot without seeing a lot of positive effects.

Like, neither of us, I think, would stand in full-throated pure defense and say these people know exactly nothing. I don't think they know exactly nothing. They have-- They're making some important points in some ways. At the same time, boy, when they say Edtech, I don't think they think about most of the listeners of this podcast, I'll tell you that.

Yes. I mean, I think they are thinking about big tech's education plays, including AI providers and large tech companies like the FANGs and whatnot. What do you make of it, Ben? I could probably talk about this 

[00:27:45] Ben Kornell: for a 

[00:27:45] Alex Sarlin: straight hour. Yeah. 

[00:27:46] Ben Kornell: Clearly, as a society, we've lost our ability to engage in nuance, and so extreme points of view cut through the noise in ways that Jared Horvath and some of the Jonathan Haidt stuff has cut through.

And it's hit a nerve that I think we in Edtech need to acknowledge, which is that screens have had negative effects on kids. And so what kind of intentionality are we as educators putting towards when to use a screen, when not to use a screen, what type of engagement, what type of pedagogic moves? I say this so much, you're probably tired of it.

Edtech is ed first, tech second. But that's not a message that's gonna be salient with this crowd, just because nuance doesn't get their point across, and I think that the realistic stance that they have is probably just reducing the role of big tech in education and in kids' lives. I think that that's the victory that they're going after.

But they've been lumping in all of Edtech in that and firing up the base with some grassroots work with parent groups that I think has ensnared educators and superintendents and also education technology providers. And so the main thing that we have to do as a space is continue to articulate the why behind our tooling.

And ultimately, so many parents who say, "We don't want any technology in our lives," also in practice, utilize technology in their lives in lots of different ways, some of which is really productive. So I, I don't think that we're gonna win the battle in the PR and headlines space, especially against somebody as good-looking as Hugh Grant.

I mean, Alex, you're very good-looking, but Hugh Grant is-- he's like, what, 65 years old, and he's still gorgeous. So we can't compete with that, but I do think that we can demonstrate that the efficacy of the tooling that we've built around education outcomes, and that's where we've gotta take the stand.

That's my take. 

[00:29:53] Alex Sarlin: I agree, and I would just add one more piece to it, and I-- and we've wrestled with this. I wrestle with this a lot, but I think it's very clear. You look at Sophie Winkleman's blog posts about this and, and, uh, you know, you read the Substacks that all of... Th-this is like a relatively small group of people with a very outsized influence because they're saying things, as you say, that are very-- In our world, they're pretty extreme.

I don't even know if they come across as pretty extreme for other people. But they have headlines like things like, "Why did we let Big Tech take over education?" And to be fair, like, that's a good headline in that at least they're defining Big Tech and not saying Edtech is bad, which is like literally the headline on the, on the Closed Screens Open Minds website.

It's like Edtech. There's a LinkedIn right now video going around from Jonathan Haidt basically saying, "What is the motivation of Edtech companies?" I really encourage everybody listening to this podcast to go check this out because it says it's asking Hugh Grant in an interview, Jonathan Haidt interviewing Hugh Grant, "What is the motive of Edtech companies?"

And he's like, "It's to collect data That's all it is. It's the same business model as any other big tech company. They-- all they wanna do is collect data and monetize. There's nothing else there. And I'm like, you know, five hundred interviews in to Edtech founders, I'm like, w-wow. I'm trying to imagine what he's envisioning when he says that.

I'm trying to imagine what person. Is it Sundar Pichai? Like, who is he envisioning when he says that the entire Edtech industry is purely for data harvesting and attention and distractions? Because it is not any of the founders, it is not any of the investors, it is not any of the operators. It is literally nobody I've ever talked to in the Edtech space thinks like that.

Nobody. It concerns me that there's people out there actively trying to paint the entire industry as the worst version of big tech. It's scary. And I guess my point is, we have to figure out as an Edtech industry, what do we want our relationship to big tech to be? 

[00:31:41] Ben Kornell: Yeah. 

[00:31:42] Alex Sarlin: What do we want it to be? I mean, 'cause you can imagine a PR firm saying, "Okay, Magic School.

Okay, Brisk. Okay, School AI. The number one thing you have to do is go around saying you're not Google, you're not OpenAI." And you have seen a deal on LinkedIn saying, "We're not Google." I can see that instinct, but at the same time, Google is a humongous and, and I think a very positive player in the Edtech space as well.

It's a really complicated thing to try to parse out, and they're making it sound like it's not complicated. 

[00:32:07] Ben Kornell: We're gonna have to leave it there. Curious what our audience thinks. Feel free to add your comments on the podcast. We're gonna be coming back to this story again and again. So sorry to wrap things up on that note, but we will be coming back to this again in the future, 'cause if it happens in Edtech, you're gonna hear about it here on Edtech Insiders.

Thanks so much. 

[00:32:28] Alex Sarlin: We have a very, very special guest on the Week in Edtech today. We are speaking with Sabine Wallis, the General Manager of NWEA, which is an absolute leader in assessment throughout the entire Edtech industry, runs the MAP assessment, among others. She joined the NWEA in December 2025, and previously held VP of product management roles with Renaissance Learning and Illuminate Education.

She's a graduate of Stanford University, where she earned her BA in economics and her master's in management science with a concentration in computer science. Sabine is a great resource for how research, assessment, and learning data can work together to better support educators and students. Sabine Wallis, welcome to Edtech Insiders.

[00:33:11] Sabine Wallis: Thank you so much. I am so excited to be here. 

[00:33:13] Alex Sarlin: I am excited to talk to you today. So the NWEA, especially the MAP assessment, is such a core benchmark for the entire K-12 industry. But we're in this crazy moment for data, right? Schools are collecting more data than ever before. They're trying to turn that data, including lots of assessment data, into meaningful data that can actually drive decisions.

What do you think about assessment data and how it can be best used to make actual changes in the classroom benefit student learning? 

[00:33:41] Sabine Wallis: Well, Alex, I think there's three problems that we need to solve. One being bringing together disparate sources of assessment data, and that might be different assessment types in some cases, it might be from different sources.

That's problem one. Problem two is being able to synthesize that data to provide instructionally relevant insights to teachers. Yeah. So not broad performance categories, not big composite scores, but rather the atomic skill that a teacher needs to focus on. So an example, it would not be that useful to tell a teacher This student is struggling with fractions.

It would be a lot more useful to say, this student is struggling with adding fractions with unlike denominators. 

[00:34:25] Ben Kornell: Hmm. 

[00:34:26] Sabine Wallis: Problem three is once you have that right level, that granular level of insight, actually providing the teacher with the action to take, so the instructional resource or strategy, but making that tangible recommendation to them.

So it's synthesizing data, providing the instructionally relevant insight, and then actually serving up for the teacher the instructional resource that they may want to use. 

[00:34:51] Alex Sarlin: When you talk about that loop of using assessment data to actually inform instruction, we have a name for that in education, it's formative assessment.

Right? And, and incredibly powerful, and there's been a movement to improve the use of assessment data in a formative sense for many years. When you think about that connected system where you're getting actionable instructional insights from data, how can we get that on a more frequent basis, so instead of having these big benchmark sort of summative exams once in a while, how do we do that in more lightweight, lower impact basis over time, reduce the amount of time spent testing, but get those instructional insights anyway?

What does that look like to you? 

[00:35:32] Sabine Wallis: That's a problem we've been thinking a lot about at NWEA, and my background before coming to NWEA is in fact in formative assessment, so I think I'm bringing that ethos to the group. But we've specifically been talking a lot about how you bring interim and formative assessment data together, right?

And assessment has always been built for purpose. Interim assessment is to measure growth typically is how we describe them, formative to drive instruction in the classroom. 

[00:35:55] Ben Kornell: Yep. 

[00:35:55] Sabine Wallis: But what we're looking at is what happens when you bring together the data from those assessments. You might find that you actually have insights you didn't realize when you were looking at those two types of assessments separately.

And those insights, we believe, can lead to reducing some of the assessments you had previously planned. So we're super excited about the direction that we're going on that front. 

[00:36:18] Alex Sarlin: I want to hear a little bit more about that because that feels like such a core direction for the assessment space in general.

And I know that one aspect of what NWEA does is think about how to have just as much meaningful data coming out with maybe less of a testing time burden on classrooms, on teachers, on students. What does that look like? 

[00:36:37] Sabine Wallis: Well, I'll speak to both, two questions in there. So MAP Growth is an adaptive assessment, right?

So its purpose is really to, as efficiently as possible, get to where our student is. So that means as items- Mm-hmm ... questions, test questions, sorry, I shouldn't use the, the item term. When questions are being presented to the student, depending on how they answer, they're gonna get served up the right question.

Is it harder? Is it easier? And that allows us to do something that wouldn't be possible on paper/pencil. It's only possible because it is a computer adaptive assessment, and we can get to as quickly as possible where that student is. In terms of some of the innovation that we have going on, I'll share a little bit.

We have a learning sciences team at NWEA that are really experts in the field, and they have a combination of what we call a learning ontology. They have a proficiency model, and that learning ontology you can think of as a graph that connects skills and standards and instructional resources. We have a proficiency model that's getting to for a student, for a class, here's their skills proficiency, here's their standards proficiency.

And then we have what we call a cascading inference model that serves up recommendations based on all of the above. But what is at its core is that we're bringing in, as I was describing earlier, these disparate data points. 

[00:37:57] Ben Kornell: Right. 

[00:37:57] Sabine Wallis: So map growth data. Right now we're also bringing in our curriculum-embedded assessment data that we have at HMH.

But we are looking at other data sources to bring in, so Really thrilled. 

[00:38:09] Alex Sarlin: Yeah, that is really exciting. And you mentioned adaptive assessment, and adaptive testing has preceded AI. It's been around a number of years, but the technology continues to improve. We've seen some assessment providers, you know, get shorter and shorter sets of questions to get the data, get the granular data, the sort of cascading inference.

I'm probably misusing that term, but that idea of getting data that can actually inform teachers about something as granular as fractions with unlike denominators out of the minimum number of questions needed. I'm curious about what that looks like for you. I know this is like we could get in the weeds for a while there, but also as that data comes out, how do you translate it, the complexity of this item theory stuff into actionable level insights that educators can just pick up and go, "Got it.

I'm gonna go do this exact activity with unlike fractions tomorrow in class for these students"? I 

[00:38:58] Sabine Wallis: think it goes back to your very first question, which is you have to present things in terms that are meaningful- Yeah ... to the teacher in the classroom. So teacher's even probably not saying, "I'm gonna instruct on this state standard today."

They're really saying, "I'm gonna focus on this skill today," or, "These kids need help with these skills." So it's providing- Right ... that skill level insight, which is what we're looking at. And then again, teacher knows best, right, first of all. So teacher knows best in the classroom. We're providing them a recommendation around a resource or strategy to use.

So they're not just given the insight and left to say, "What do I do with this information?" We're actually giving them to help them institute that insight. 

[00:39:35] Alex Sarlin: Yeah. And you mentioned earlier that HMH and NWEA are really under the same umbrella now, and you can use different types of assessments, curriculum-embedded assessments from HMH, MAP Growth assessments from NWEA all together.

That seems incredibly powerful, and we've talked to some of the HMH people about their strategies, really creative and interesting strategies for how they're putting all the pieces together. I'm curious what that's looked like. Tell me how the pieces fit together under this large umbrella of a major Edtech publisher.

[00:40:04] Sabine Wallis: Yeah. So when I described how do we make data more actionable, I think one of the problems we've always had in the industry is that the pieces didn't all live under one roof, making it really hard to actually build towards that vision. So the beauty, one of the reasons I came to HMH to take this role was we have all of those pieces under one roof.

We can collaborate closely, and so ultimately we can create those connections between MAP Growth, really all of the data we're collecting, and those resources on the HMH side. Yeah. But I think it's worth pointing out that as NWEA, it is important to us that we are working with-- We understand that districts work with many different curricula, or in many cases they might have HMH for something and another provider for something else.

So part of our promise is that we are, at the end of the day, curriculum agnostic. We want to be able to support different programs and providers, and we have tools towards that. But of course, we're gonna be able to do the best work with our HMH colleagues because we're all in one room. 

[00:41:08] Alex Sarlin: It makes sense, and you can easily imagine how having all of these teacher guides, all of these practice questions, all of these lesson plans, all of these curriculum items at your fingertips can make that transition from data to action much smoother than it has been in the past.

But as you say, it has to be agnostic. It can't only work with districts that are already using that curriculum. We're in this era of evidence-based AI. I think it's, it's on everybody's mind. We-- Just today, we saw a big letter from the Department of Ed really asking everybody to focus on impact and focus on evidence.

And one of the ways that that happens in the Edtech space, as you know very well, is people do research to sort of validate products and make sure they work. That is incredibly important, but you've said that, you know, research should go beyond just basic validation of products. What kinds of research do you think have the greatest potential to actually move the needle on student outcomes and support data-driven decision-making in the Edtech space?

I 

[00:42:01] Sabine Wallis: think like you said, number one thing we have to do is understand, is this particular product or tool going to have the impact that it is claiming to make, right? That's number one, and we do think MAP Growth has an important role to play there- Yes ... as an independent arbiter of growth. But I think even more meaningful is understanding where that growth comes from.

So what are the particular instructional practices or strategies or interventions that are driving that student growth? 

[00:42:32] Ben Kornell: Hmm. 

[00:42:33] Sabine Wallis: And then asking, well, in what context and for which students, right? In what classroom context and maybe different for different Groups of students. And once we have that deeper knowledge than just this product works, educators can make the right tools around their-- ma- right decisions around their investments, but they also have the information to really drive decision-making in the classroom.

[00:42:57] Alex Sarlin: That's really interesting. And, and to being able to sort of segment different types of interventions against different groups or people with different prior knowledge. It's not just this product works and this product works, but this product doesn't work. It's when do they work? For whom do they work? In what context?

With what dosage? We all wanna get to a more granular understanding of that. I think the entire field wants that. And as you mentioned, you know, the MAP Growth has played a very special role, I think, you know, for a decade or more as being independent arbiter, as being sort of this clockwork assessment that whenever people are testing different interventions, whether they're tech-based or not, you can look at the MAP Growth as the core metric for what is for the outcome.

I, I'm curious what that looks like from within the MAP world. Do p-- do Edtech vendors, do instructional coaches, do, do, uh, people who are, who are coming up with new instructional, you know, models come to you and say, "Okay, we found something that can improve MAP Growth. How would we work together?" I'm just curious.

Or are you just totally independent? 

[00:43:52] Sabine Wallis: All of the above, right? Yeah. Folks reach out. Maybe one thing I'll mention is we have something called the Instructional Connections Program within NWEA, and we work with thirty, about thirty different instructional providers, and partly to do things like place students into their programs, but also so that within those programs, there can be this independent measure of growth.

Right. Is this program having the impact? So, you know, researchers and different products and school districts, all of the above, sometimes working with us, sometimes independently. 

[00:44:24] Alex Sarlin: That's really interesting. It's called the Connections... 

[00:44:26] Sabine Wallis: The Instructional-- ICP is the acronym, Instructional Connections Program.

[00:44:30] Alex Sarlin: I've never heard of that, but that sounds really intriguing. It sounds like an exciting sort of working group of different people coming from different approaches, but all with the same shared goal of improving outcomes. So another sort of hot topic and a sort of perennial hot topic in Edtech is the concept of integration or interoperability, sort of how all the pieces fit together into a sort of stack in the actual classroom space.

It's easy to talk about that. It's very hard, especially for individual districts, to sort of put all the pieces together, make sure they're all talking. In the HMH or the NWEA world, what should integration look like, and how, how do you think about the future of it? 

[00:45:05] Sabine Wallis: As you can imagine, we talk about this a lot, and I think we have some fundamental principles about how we, we think about this, which is that integration should start in a teacher's workflow.

Meaning- 

[00:45:16] Alex Sarlin: Hmm ... 

[00:45:17] Sabine Wallis: we can't ask teachers to just wholly change the way that they are doing things in their classroom. So it's not enough to just have the different ingredients to some integration recipe. We actually have to meet the teacher where, where they are. Hmm. And that means if there's some new resource or new insight or new report, what have you, that we wanna serve up, it has to happen At the moment that the teacher expects it, which I don't think is how everyone approaches this problem.

And then I think the other piece is keep it simple. I think too often integrations add complexity to the educator's workflow. Mm-hmm. Where at the end of the day, any integration should be making things easier for the teacher, making things easier for the educator so that they can do the work that they do best, which is engaging with and connecting with and working with students.

I think the other piece is just always having our eye on a quality bar and making sure that we're not, you know, trying to do something quickly or shoehorn things together, but we really need to provide educators with the quality of product that they deserve. 

[00:46:19] Alex Sarlin: That makes a lot of sense, and I, I love the idea of, you know, making sure to meet the teachers at the moment of need, 'cause I think, you know, every educator knows that they have many different Edtech products.

They all have reports and dashboards somewhere, and you can access them, and you, you know. But is it an everyday thing? Is it once a month? Is it quarterly? Is it at the time of the interim assessment? So the idea of actually having, you know, making that decision easy for them and saying, "Here's something you're gonna need right now."

It'll, has very specific recommendations or ideas of what you might do with it. It's not just data for its own sake. That feels like it should definitely move things in the right direction. I'm curious, you know, this is just a very broad brush question, but y- N- NWEA i- is such an interesting organization, and the MAP exam really took flight over the last decade and a half maybe or so as becoming really one of the absolute core benchmark assessments or interim assessments in education.

I'm curious what it feels like, you know, what you see your role as an, at NWEA helping the entire field uplevel. We know we've talked about a few ways in which that happens here, but I'm curious just zooming all the way out, like, what would be your dream about saying, "Hey, with NWEA allowed this to happen, and it changed everything for classrooms"?

[00:47:32] Sabine Wallis: I'll start by saying I think NWEA has always been goal standard and measurement, and that's because of the scientific credibility, the validity, and reliability behind the assessment, and that's still central to who we are and who we wanna be going forward, right? Being that trusted measurement and creating something with rigor.

But I actually think there's a lot of opportunity to do more, and it's, it's all the things we've talked about to some extent, right? It is, uh, being able to leverage that assessment not just for growth but to drive instruction more effectively, which also means ultimately connecting those instructional insights to the activities in the classroom.

So I'm excited to preserve a lot of the philosophy that NWEA was founded on but also to lean into some new innovations. 

[00:48:22] Alex Sarlin: Yeah. We-- You said you, your background in formative assessment and that, that resonates very well. I-- Willem and Black and all those great folks, I, I love that body of research incredibly.

That-- It was one of the real eye-opening as an instructional designer, I feel like the formative/summative assessment dichotomy was one of the real eye-opening moments for me about how education works, and I, I love hearing you say it. You're leaning into that. We have managed to go this whole time without mentioning the word AI.

I don't even have a question here. I just want-- I'm curious what-- how you're thinking about AI in-- from the NWEA map perspective or NWEA perspective. 

[00:48:56] Sabine Wallis: Yeah. I mean, I think we are primarily focused on are there ways that we can leverage AI to enable the teacher? A good example would be the explainability of some of our insights, but really our focus is on the teacher, the educator use case.

But, you know, why do you see this trend on your reports, et cetera. Those are the sorts of things that, that we are looking at. But I think some of the trust and transparency I talked about before around making sure your product has impact. Because of all of the emerging products and tools in the market happening right now, one of the reasons that we're really emphasizing this trust piece is to make sure that the tools that do come out that might leverage AI capabilities do have the impact that, that- Right

educators and students deserve. 

[00:49:45] Alex Sarlin: That is a very valuable and very needed role in the market right now as all of these AI products just appear and, and every incumbent, including HMH, starts incorporating AI, AI into their products. We all wanna know what's working and what's leading to growth. Makes a lot of sense.

Thank you so much. This has been really fascinating. Sabine Wallis is General Manager at NWEA, and she previously held VP of Product Management roles at Renaissance Learning and Illuminate Education. NWEA is, of course, part of Houghton Mifflin Harcourt, HMH. Really, really loved to speak to you, so thank you so much for being here with us.

[00:50:17] Sabine Wallis: You too. Bye-bye. 

[00:50:18] Alex Sarlin: Bye. 

[00:51:23] Ben Kornell: Hello, EdTech Insider listeners. I am so excited to be joined today by Andy Rotherham and Marissa Mission from Bellwether. Andy is the co-founder and senior partner at Bellwether. It's a national nonprofit. I've actually worked with Bellwether back in the day at Envision Learning Partners. Bellwether was a great help for us navigating the '09 housing crisis.

That really feels like the glory days when we look at the challenges today. Andy was a former White House aide, Virginia Board of Education member, education analyst and writer, and author of the Eduwonk.com blog and newsletter. So he comes from a family of blogger and newsletters. Welcome back to the fold, Andy.

And Marissa is a senior analyst at Bellwether, where she translates complex research and policy topics into actionable insights for educators, leaders, and policymakers. And much of her work is focused on AI in education, which we'll be talking about today, including tinkering with Bellwether's workflows, as well as investigating AI's effect on productive struggle and how technical design choices can lead to better learning outcomes.

Can't wait to dive in. Andy and Marissa, welcome to the pod. 

Andy Rotherham: Good to see you. Thanks for having us, Ben. 

Marisa Mission: Yeah. Excited to be 

Ben Kornell: here. Before we dive into AI in education, maybe Andy, can you just tell us a little bit about the journey for Bellwether, and where did it start and what you all do today? 

Andy Rotherham: Yeah. So thanks for asking.

It's been long enough, people don't always ask about that anymore. And so yeah, you were working with us back in two thousand and nine. That's when we were just getting going. So there was four co-founders at the time, and we sort of brought different areas of expertise: strategy consulting, recruiting and talent, sort of innovation and policy analysis, and research.

We brought that together, and the idea was just to build a firm that operated in certain ways that we thought were good ways to operate a nonprofit business in the sector. But also, we're laser-like focused on this issue of sort of capacity and conditions, so helping build capacity inside various kinds of education organizations and school districts and schools, and trying to do what we could to create context and conditions that are more likely to lead to effectiveness.

And we've had some twists and turns since then, obviously. We've grown. There's about a little more than 90 Bellwetherians now all over the country. We were virtual from the jump. We have offices in a number of cities, but we've operated nationally and virtually. We're still doing those kinds of work.

We've also expanded into a number of other things. We do a lot of academic work in schools now, and a lot of hands-on work. There are a lot more implementation work, things like that, than we used to. But that sort of core DNA of helping people solve problems and in trying to improve the context and conditions they work in, that's the same 

Ben Kornell: It's incredible, and I saw a bunch of promotions coming across, too.

Seems like there's a real community within Bellwether that's now formed, almost like having been on the for-profit side of the ledger for many years, like there's this Bain culture that they created at Bain where then you actually find that it creates this ecosystem of talent itself. I mean, there's this Bellwether talent hub with its own diaspora that's just incredibly powerful.

So congrats on all of that. And there's a million things we could talk about today. I would encourage folks to go to your news and commentary page, where there's just always great stuff that I find there in dipping into finance and dipping into enrollment declines and so on and so forth. But today, Marisa, we have you here to talk about AI.

And I 

Andy Rotherham: have to note, Ben, Marisa recently promoted, so she's a real talent on our team, so apropos of where you're going with that. 

Ben Kornell: Oh, congratulations, Marisa. What's the new title? 

Marisa Mission: Thank you, Ben. Thank you, Andy. I am now an associate partner at Bellwether. 

Ben Kornell: Associate partner. Wow. Well, I'm gonna have to redo the intro there.

That's great. 

Marisa Mission: All good. 

Ben Kornell: Well, I think everyone is talking about AI in education, what's real, what's not, what's efficacious, what's not. We just did a research series where it was showing that cognitive lift was declining, and yet we're also really faced with procurement processes where districts are trying to buy products that by the time they implement, they don't even know what the product might look like 'cause it's changing so fast.

So can you give us a little bit of the state of play of AI in education and the research that you've been doing? 

Marisa Mission: Of course. Yeah. I think it's true that districts and educators really have been absolutely inundated by this flood of AI tools. We hear it both from the education side of the world, but also the tech and startup side of the world, where VC capital is really interested in not just AI startups in general, but specifically AI education use cases, so for example, AI tutoring.

In fact, I just recently heard that the Y Combinator has a specific focus on AI tutoring, and the majority of its newest class is focused there. And so we know that a lot of folks are interested in what AI can do for education. But it's a lot harder for the folks on the education side to really think about not only their day-to-day of running schools and educating students but also where technology and this new version of technology lives in that ecosystem.

And so as you mentioned, Ben, it can be quite difficult for folks to set aside time to investigate what effect implementing a new AI EdTech tool might have on a classroom or on a student. 

Ben Kornell: So let me double-click on that, Marisa. Part of your work at Bellwether is the meaning-making of, okay, how do we think about this?

What kind of advice are you giving to both district side, but also EdTech side to help clear the fog of confusion? 

Marisa Mission: Yeah. Over the last couple of years, we've really dug in on both cognitive science research side, but also talking to folks in the field, and that includes both educators as well as developers.

And through all of that research and conversations, we've come up with what we think are five signals of quality that we've written about and think will be really helpful for folks as they evaluate whether an EdTech tool is actually useful. And there's three that I think most folks could come up with on their own.

They're all related to instruction and measuring outcomes. But there's two that I think that folks don't necessarily think about first when it comes to this. One of those is technical design, and a lot of my work internally at Bellwether is thinking about how AI can be used properly. And so not just in a safe way, but really increasing the quality of the output.

And often that means digging into some of the technical configurations around prompt engineering, around cleaning data, around really just which AI models to use, for example, and how to fine-tune output. And I think a lot of folks in our sector tend to brush that off as IT concerns or really technical concepts, when in reality, they're really, really key to how well AI is able to, for example, re-level a text for a second grader versus a fifth grader.

How do you know when it's good and when the AI has done it well? Well, what levers are you gonna push? It's in the technical configuration that you're going to make it better. So I think that's something that folks really have to pay attention to And then the second thing that I would really highlight is actually nothing to do with, uh, technology at all, but more about market sustainability.

So we have a bunch of these EdTech startups that are coming out of the woodwork, right? But we also have these massive tech giants that are getting into education. So thinking about Anthropic, OpenAI, Google. Google Classroom is already such a ubiquitous tool, but Claude for Teachers just came out a couple of weeks ago, and ChatGPT for Education has-- I think they recently added maybe five or six new skills and plugins.

So there's a new focus and a renewed, I think, vigor coming from these big tech companies, and it's going to be a question of what works better in a school in the context of a teacher who is going about their every day. Is it going to be an EdTech tool that is maybe smaller, more curated, or is it going to be a frontier tool that can just be easily adapted to a specific situation?

And so these are a couple of the things that I don't think as many folks are really paying attention to, but would be really key for something like procurement 

Ben Kornell: Yeah, totally. I mean, Andy, as you've seen the broader space shift, the large tech players, and The New York Times points this out, Natasha Singer has a great article around big tech in K-12 schools.

There's a way in which that the big tech can crowd out things, but there's also a way in which educators can now be builders. So sometimes edtech providers are caught in between. So when you're looking at how you think the space is going to play out, whether it's gonna be homegrown tools in schools and districts, whether it's gonna be big tech tools, or whether there's a thriving space for edtech, how do you see that space playing out strategically, and how are you advising your partners to navigate that?

Andy Rotherham: Yeah, I think you're onto something, and I do think there's a whole set of builders out there. A lot of them are under the radar. They're the teachers of my generation. They're the teachers who had sort of the Apple IIe and were doing stuff with that. They're that same kind of teacher now with today's technology, and they're building tools out of AI.

I also just think what's happening with the tools that are coming out, what you can build, if you're not gonna win a tool-based arms race in the education market, so your question is exactly the right one. What are those differentiators? And I mean, schools struggle with change management, and we see some of them right now, some of the wealthiest school districts in the country are still struggling with supporting their teachers, implementing AI effectively, giving their teachers the training and support, that human function around that.

And so it's not just the tools, it's what is your ability to actually come in, help with effective implementation, change management in that human dimension. That's gonna ultimately be the real differentiator, and that's the place where we're advising people to make sure you pay attention to because if it's just the tool itself, you are constantly going to be at risk, not only from your competitors, but as you said, from people who can just come along on top of the stack and build pretty amazing stuff.

And I think we're just sort of starting to see some of the stuff that teachers are building. I think you're gonna see a lot more of that 

Ben Kornell: So one of the things that, that Bellwether has covered also is the rise of education savings accounts and education delivered outside of schools as well. So really outside of the school building or outside of the school day, where are you seeing high potential for effective AI usage in education tools?

Do you think that the ESA markets are sustainable, growing? Do you think that that's a market that edtech should focus on? How do you see this space evolving with that? 

Andy Rotherham: Yeah, this actually is something Marisa and I have talked about. Be interested to hear her take. Mine is, yeah, you have to focus on that.

These are gonna be large revenue streams. There's plenty of problems with them for sure, and I think you're gonna see some backlash, and there's gonna be calls for accountability and so forth. But this general unbundling that we're seeing of how public education is delivered, that's not gonna go away.

It's gonna get modified. You're gonna again see correctives and so forth. But that's the trend it's going, and AI will be a piece of that insofar as we talk about sort of tutoring and supplemental activities, things for kids out of schools, and ways kids are accessing and managing their learning. So it's gonna continue to be an evolving and robust market.

But Marisa, as I said, we haven't-- This actually isn't one we've kicked around. I'm, I'm curious to hear your take. 

Marisa Mission: Yeah. I think there is the current landscape, and then there is the potential future. And in the current landscape, we're actually not seeing AI be implemented very comprehensively across the OST landscape, whether that is extracurricular or non-traditional schooling at all.

And so one of the biggest takeaways from some research that a few colleagues and I recently did is that there are some, I think, boundaries or, or barriers to implementing AI effectively, partially because it is such a inconsistent space. You never quite know if an AI tool is being used in the same way across different students, different classrooms, in places like a micro school or the micro schooling sector and in, in places like the OST sector.

And so currently what we have is quite fragmented, and that in and of itself presents something of a barrier because there's a large benefit to economies of scale when it comes to technology. And non-traditional and the OST sectors don't quite have that benefit the way that districts do. 

Ben Kornell: Yeah. I mean, Marisa, by the way, just to jump in on that, um, you know, one of your point, uh, one of your five areas is sound pedagogy and coherence with existing instructional practice.

And within a school district, you can go for coherence. I think we're experiencing a lot of incoherence in, you know, homeschool and ESA sector space. And so then that addition of AI and AI tools just adds to that fragmentation. But go ahead. 

Marisa Mission: Exactly. In-- If we think about the potential for AI, there is a huge amount of potential for ways to help students tailor their, for example, homeschooling materials to their level, whether that is something easier or harder, whether a parent is an expert in a field or a particular topic or not, students can get what they need, and that is honestly what every parent wants for their kids.

But on the other hand, if tools are not necessarily high quality or they're not built on high quality standards, it becomes a game with winners and losers, where the folks who happen to choose the right tool get these amazing benefits, and the folks who didn't are maybe even worse off than if they hadn't chosen the tool.

Mm-hmm. And so I think one of the things to pay attention to outside of the traditional school, public school system is really thinking about how we incentivize the market to make sure that these tools are going to be either high quality in and of themselves or built on really high quality materials. 

Andy Rotherham: I do think we should-- like, a couple of just things to know.

One, the direct consumer market is always the trickiest market, and so folks are gonna continue- Mm-hmm ... to focus on where kids are aggregated, which is in schools and so forth. But some of these things are distinct in the sense of parents, like homeschooling parents, they like point solutions. Like, there was an author who referred to, you know, they'd go to sort of various kinds of online models to sort of a stage of grieving, and then they realized that's actually not what, what they wanted.

They wanted the real customization, and so they end up doing that themselves with various point solutions. That's much more of a challenge in school, the coherence thing you talked about a second ago, Ben. And so these are sort of distinct marketplaces that are looking for different things, but that revenue source is gonna continue to grow, and I think you will see some activity around it.

Ben Kornell: Okay. So the last area I wanna dive into is apropos for all of these surfaces, both the big tech, general AI, the in-school, you know, adaptive learning programs, and the homeschool market is all coming down to productive struggle. And this is something that you've really highlighted. You talk in your article a little bit about Quill and Edlight, and of course, everyone wants to drink their own Kool-Aid.

"We've got productive struggle." But what does it really mean and look like, and how could you actually You know, implement processes where you're ensuring that pr- productive struggle is really happening, whether it's a parent, a child, or an educator. That seems like the holy grail here. Can you tell us about that?

Andy Rotherham: Yeah. It's one of these words that means a lot of things to a lot of different people and in different contexts. So people talk about it in terms of math instruction, a slightly different context. What we're talking about here is just learning, like the act, that actual act of learning. And AI can be a fantastic expediter for a lot of things.

People-- I mean, I'm sure we all use it in various ways professionally, but you've got to know how to do those things first. And then I've likened it to carpentry. A, a really skilled carpenter can take a lot of shortcuts and still get to a good result because they're a skilled carpenter. They've learned.

They didn't learn doing the shortcuts. We have a huge challenge with helping teachers teach like that, and we haven't been supporting that. That's a piece of the change management. How do you actually teach in a way that encourages this? And they're not getting a lot of support on that. Helping students and designing experiences for students so they understand the importance of this, 'cause what we're seeing right now is kids are jumping to the solution.

That's what is called cognitive offloading. So essentially, these, these are fancy words. Productive struggle basically means learning. Cognitive offloading basically means skipping that learning. And we need to-- You know, and kids, as we know, they put a huge discount rate on everything. So you go tell them, "You shouldn't do this 'cause it's gonna have a consequence down the road," and they're like, "Okay," you know, you know, "Get out of here, old man."

So but helping them understand that you do actually need to do that work, and there's just not a lot of change management. Instead, what's on offer is mostly, "Okay, let's just ban this stuff," which we can argue about sort of age appropriateness and, like, do you want young kids with agentic AI? There's, like, important arguments and conversations to have there.

But the idea, like when you're at the high school level, that you can sort of ban your way out of this, it's not on offer. This is not going away, and so we need to figure out, okay, how do you train and support teachers to actually teach effectively in this new environment? I think that's a place where we're just falling down.

Teachers are kind of being thrown to the wolves, and that's fueling these calls for bans and so forth, and that's all very shortsighted. 

Ben Kornell: Yeah. So Marissa, can you tell us about some of the positive like programs or platforms that you've seen where productive struggle is really enacted? 

Marisa Mission: Yeah. And this is where, as I mentioned earlier, technical design really comes into play.

Productive struggle is really about adding friction, and that could be for anyone, anywhere, whether that's a student or a teacher, administrator. And so the question comes down to where is it appropriate to add friction and how much friction can or should we add? There is a point we know where the struggle becomes unproductive and students become unmotivated or demotivated, and they lose confidence in themselves, and that's definitely not what we want.

But we also are trying desperately to avoid cheating and the jumping straight to a solution where there's no friction at all. And so what really has stood out to me about certain tools is the way that they thoughtfully add friction. I'll give you an example using one of the organizations that we did a case study on, Edlight, and where they add friction is actually in the solving of math problems by hand.

So students cannot actually use the AI tool until they have taken a crack at it by hand. From there, what the AI does is evaluate their work, identify misconceptions, and then provide feedback to either the student or the teacher, depending on how the administration has it configured. And so in this particular sense, there is friction to be added in a student completing a problem on paper.

There is also friction that could be added in whether or not the feedback is given to the student or the teacher. If it's given to the teacher, there is an additional check that the teacher has to do of digesting that information and then thinking about how to present it back to the student, for example.

And in certain classrooms, that might be desired for new teachers, for example. We know that teachers in years one to three really need those reps and that practice But for a veteran teacher who maybe has 35, 40 kids in their classroom, that might not be the best way of moving forward. And so for me, productive struggle is really about thinking critically on where it makes sense to add friction.

Andy Rotherham: And one thing that Marissa alluded to, and then we, we kind of see is also for teachers, productive struggle for teachers and cognitive offloading and like what we're seeing, and one of the things AI can do is essentially make processes that aren't great slightly more efficient. And so one of those is like, I think a lot of people don't like teachers grabbing lesson plans off the internet, off of Pinterest or whatever that aren't high quality or aligned.

Well, now you can, you know, ask ChatGPT or Claude to build a lesson for you. It may or may not be aligned or, or be of quality, but if you're just implementing in the same way you're implementing the one you used to just grab offline, that's teacher offloading, and that's not helpful either. And there's, so there's a whole, there's a whole separate conversation to be had there.

Marisa Mission: Andy, that example is, is perfect too because of the recent EdTech tools that have been launched inside of ChatGPT and Claude, for example. So when I talk about technical design and the way that can be used to improve the output of an educational product, Claude for Teachers really at its core is a set of agent skills that can do exactly what Andy mentioned and create a lesson plan.

What you're going to get out of using Claude for Teachers is going to be vastly different than if you just used normal Claude or normal ChatGPT. And so this is one of those areas where if we focus on productive struggle and thinking about friction for teachers, I haven't used the skills yet them myself, but a skill could add friction even as it makes it easier to maybe write the entire output.

It can add friction by asking teachers to think through, what are the goals for this lesson plan? How does this connect to your standards? By asking teachers to do that upfront work, even as Claude is making it easier to put the final thing together, we can add some productive struggle for them. 

Ben Kornell: Yeah, it's interesting the people that are using Claude, you know, Claude has kind of a universe of partners that they've worked with to build integrations with.

And one thing that's interesting about the design is instead of one-shotting outputs, it actually slows the process down, which, you know, my thought initially was that that's in order to make a better end product. But now hearing what you're saying, it is also that you want to have productive struggle for the teacher and as they're thinking about the lesson plans or the materials and so on.

So that's a great insight. And you know, just today, OpenAI announced fifty-five new school districts that have signed contracts with OpenAI to be their AI provider. So we are seeing that intrusion or expansion of big tech into the K-twelve ecosystem more and more. We're gonna have to leave it here, but if people wanna find out more either about this work on Five Keys or around productive struggle, where should they go, and what's the best way to learn more about what Bellwether does?

Andy Rotherham: They should come to our website. If you just come to the website, you'll see, first of all, there's an opportunity to sign up for our newsletter, and you can see some of the different issue areas we work on. There's a whole landing page specifically on AI where you can learn about some of the work we talked about here and a lot of other work that we're doing.

You can also just reach out. I'm andy@bellwether.org, so we're very easy to find. Marissa, her, her name's as marissa.mission@bellwether.org. Just reach out. We'd love to connect. You mentioned Eduwonk. You can find a lot about what we're up to in various Eduwonk posts at eduwonk.com. And we have an AI-specific newsletter that if-- I suspect for your listeners would be the one that they'd also wanna go ahead and sign up for.

Ben Kornell: And a lot of times we're including links to those also in our newsletter. So thank you so much, Andy, Marissa, it's been awesome. We could talk for hours, but for those of you that are listening, it's bellwether.org, B-E-L-L-W-E-T-H-E-R.org. Thanks so much for joining us today on EdTech Insiders. 

Andy Rotherham: Ben, thanks for having us.

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.