One role I have come to value more in moments of technological and institutional upheaval is the narrative steward.
A narrative steward helps a community make sense of where it has been, what is changing now, and which questions will matter next. In K-12 innovation, Ben Kornell is one of those people. He is darn good at it.
That role feels especially useful right now, because AI is forcing education to revisit old arguments under new conditions. Personalized learning is coming back. So is competency-based education. So is the promise of school models built around projects, work, mastery, and student agency. The ideas are not new. What may be new is the possibility of actually making them work at scale.
Ben has earned his perspective from many angles. He has been a teacher, a school board member, an edtech CEO, an investor, and one of the closest chroniclers of the sector through EdTech Insiders. His views come from living inside schools, companies, markets, and policy conversations. That gives him a kind of pattern recognition you do not get from watching from a safe perch.
In this conversation, I ask Ben what K-12 should learn before it repeats past mistakes. If we are about to relitigate “personalized learning,” what did the last decade get wrong? If ESAs and new funding models unbundle public education, what do they make possible, and what risks do they expose? If we were designing an AI-native high school from scratch, would it look like a school at all?
We also spend a lot of time on assessment, which Ben and I both believe should become the backbone of school innovation over the next decade. For years, educators have talked about critical thinking, collaboration, problem solving, argumentation, creativity, and learning how to learn. AI makes those capacities more urgent. It may also make them measurable, visible, and actionable in ways that were previously too hard to scale.
This conversation is Ben as narrative steward. I hope it helps you see the next era of K-12 with more clarity.
-Allison
On What New EdTech Builders Need to Know About Today’s Market
ALLISON: What’s something you’ve changed your mind about in the last year?
BEN: Originally, I thought that AI would spur the next set of unicorn companies across every industry, including education. That was in the period where new AI native learning platforms like Magic School, Brisk, and School AI were emerging; they accumulated a lot of capital and they were growing really fast.
Now that we’re in a later inning of AI, I’m seeing how much can be built by micro teams. The new ecosystem is probably going to look much more like a barbell—there will be tons of very, very small companies with 10 million ARR and 10 people, and a few large companies on the other end. Their primary advantage will be distribution, not innovation. They will be able to fast follow anything interesting in the microcompany space and add it on to their distribution channel. That stands in contrast to today’s market, which has more of a bulging middle. I think this will change in the next five years.
That will have massive implications for how companies build, including what the capital stack and headcount needs to look like. It’s probably the most exciting time to build ever.
ALLISON: If you’re a founder today, wanting to build something at the intersection of AI and K-12, what are your choices? How are you thinking about your capital stack, your ambition, your go to market? And then as a follow-up: what new, different categories will founders start to fit themselves into?
BEN: One: pre-product companies are a thing of the past. It’s so easy to get a V1 up and running, that you should have a product before you have a company.
Two: defensibility looks different today. Builders can go after much smaller total adjustable market sizes and still have a really nice profitable company. That’s a big opportunity for Edtech, I think.
At the same time, it’s never been easier to copycat, so once you find success, know that user experience and delivery will be your moat, not necessarily your technology. For K12 entrepreneurs, that looks like really knowing your customer segment and standing up those use cases that drive impact. That’s where differentiation exists.
Three: AI is a horizontal technology. It’s great at a wide variety of things. So builders should also be thinking about what impacts their hiring and purchasing. There is so much opportunity to learn from other founders about how to build as lean as possible with AI.
On The Future of K-12 — Priorities, Mindsets, Modalities, and Assessments
ALLISON: What do people systemically misunderstand about K-12’s current moment?
BEN: There is a lot of entrepreneurial and investor energy going into how to make the existing system more efficient, to make it work better for teachers and students. Ultimately, that’s a low aspiration. There are so many more ways to assess beyond the high-stakes, multiple choice testing format—which is the foundation for our entire system—and we’re leaving so much learning on the table by playing by these old rules.
ALLISON: I think about this story all the time: the first ever TV show was just two guys behind a table talking into microphones—so, functionally a radio show that they then broadcast on TV. They took the old form factor and then they transplanted it into the new form factor. It took many, many years to start to take advantage of the affordances of the visual media of television that then led to all the visual media that we experience today. It feels like we are at the beginning of a similar transition in how we understand and use the affordances of AI.
BEN: I agree. And it feels quite urgent in the context of education, because we’re not preparing kids to go to work today. We’re preparing kids to go to work 15 years from now. So if we aren’t thinking through how to redesign education now—which also brings up those existential questions about the purpose of school, who gets educated and why, and what do students and families value—then we aren’t delivering an education that serves them.
My guess is that K12 will be a lagging sector because the childcare value prop is so high that we’ll be willing to slog through. But the velocity of change in the adult space is moving so fast that eventually, it will have to meet that moment one way or another.
ALLISON: How do we actually map the needs of our workforce 10, 15, 20 years from now to what happens in school today?
BEN: There’s good news here. For one thing, we’re going back to basics in terms of the skills we want students to acquire. Critical thinking, collaboration, problem solving, articulating an argument, knowing how to learn—all the competency-based metrics we’ve talked about for 20 years are actually super critical. We aren’t going to know what the jobs of the future will be, so aligning to them is an impossible task. But these skills are what will allow kids to thrive in a world where their job and industry will change repeatedly. And educators already know a lot about how to teach them. What we can’t keep doing is making school about checking boxes to get to the next level.
On practical advice, I’d encourage educators to create innovation zones within their schools. That could be a grade-level team, or one school in your portfolio of fifteen, or a house system where a third of your students are in an innovative set of classes. But you need to carve out space to actually do the work of transitioning to a modern education stack that helps kids thrive. That’s how you figure out what works and what doesn’t.
Like healthcare, education delivery is intensely local—we can provide 80% of the high-quality curricular resources, but the last 20% is up to schools to adapt to their context and their students. I’m optimistic about this too, because AI can enable schools and teachers to build solutions for themselves.
ALLISON: I want to stay on the topic of future skills and competencies. We’ve long had many, many different names to describe what you call the basics: the power skills, the durable skills, the human skills. They include problem solving, working through ambiguity, persuasion, creativity, et cetera. They’ve always been important—in my view, AI makes them existential. But AI also gives us the modality to assess them at scale for perhaps the first time ever, in a way that’s reliable and verified. It’s a really exciting moment for assessment, and I know you agree.
What’s your big, bold vision for assessment over the next five years?
BEN: I’ll start by outlining what assessment feels like for students today. Just last week, my two boys (10 and 14) finished up two weeks of state testing. These are the grueling half-days of testing, after which no learning gets done because the kids are exhausted from the morning. We’ll get those results in October, when they’re already in a different class and the tests are totally irrelevant. For my boys, I’ll get a report saying they’re in the 99th percentile and there’s really nothing to work on—so it’s entirely unactionable. And this testing regime was built this way simply because of how hard it was to instrument testing at scale.
Here’s how I’d like it to look in five years, once we’ve become fluent in the affordances of AI:
Every Friday, I—the teacher—do an eight-question check-in with my students. That data is aggregated, and over the weekend I get a report: here are the skills you need to reteach, here’s the zone of proximal development for these students. My curriculum receives recommendations about the curricular or instructional moves I could make in the coming week. That data aggregates up to dashboards at the administrator level, and even the state level, where leaders can understand learning gaps by competency and by standard in near real time. It’s a nice vision—but the truly transformational one takes it a click deeper, moving away from knowledge-based standards and toward competency-based ones.
The desire for this kind of assessment isn’t new. In 2016, Education Reimagined put forward a thesis for learner-centered education that described a similar from-to scenario. Summative assessment has been around since the 1970s. But pre-AI, implementing learner-centered education burned out teachers and overwhelmed students—we didn’t have the data to make thoughtful connections. Today we have the tooling to meet the moment.
ALLISON: What’s the biggest bottleneck in achieving this vision?
BEN: Like everything in education, there’s like a chomping motion here, where the top and bottom both meet to cut off progress. From the top, policymakers, states, and districts need to set new standards and accountability systems.
From the bottom, teacher practices need to shift to data-informed pedagogic decision making. That will take real work, and it needs to be the ultimate directive across teacher professional development.
Finally, we need to focus on raising the ceiling, not just the floor. Our system today is almost exclusively oriented on remediation. When I was on the school board in San Carlos, the district had 96% proficiency across the board in reading and 89% or 90% proficiency in math. Parents were thrilled. But the entirety of both programs was built around intervention to help students meet the minimum bar of the state. That looks like being able to do eighth grade math upon high school graduation. Every parent I know wants more than that for their kids.
Policy needs to do this work. The Mastery Transcript Consortium put out a framework for a continuum that allowed students to progress without a ceiling—once a third grader mastered third grade standards, they could move on to fourth grade, then fifth. The thinking has been done. It’s up to us to actually implement it. AI can help, by translating standards across state paradigms—a long-standing roadblock to change. Think about how useful it would be to be able to generate a students’ California state standard report card, or Montessori scorecard, or an international baccalaureate transcript at the push of a button.
The only way to realize it, though, is to just get started moving in this direction. Paralysis is the biggest roadblock in schools today: they just don’t know where to start, or they’re worried they will get in trouble by a regulatory body. But the system isn’t working for most students. And families who aren’t being served will be thrilled by the opportunity to opt-in to something different. It’s also how we’ll create the demand for the rest of the system to change.
On Today’s Definition of Personalized Learning
ALLISON: I think the field is about to relitigate “personalized learning” under the banner of AI. You obviously lived through the last wave of renegotiation in a pretty intensive, hands-on way. What did the last era get wrong about personalized learning? And how do you expect this definition to change this time around?
BEN: In the last wave, personalized learning was largely a surface-layer adjustment premised on speed and engagement. One model created opportunities for students to go faster or slower based on a learner’s ability. It used assessment to understand where students were and where they should go next, but it was a very linear framework. Another model zeroed in on making content more interesting for individual learners: some kids like learning via pizza metaphor, others baseball. But in focusing on speed and engagement, those systems never actually reimagined competencies.
There was also an overfocus on individual learning rather than group and collective learning. That misses the point of how kids actually learn, which is vastly more social than people think.
Alpha School is a modern inheritor of that kind of personalized learning. You do the two hours of AI-enabled learning, you go to the learning gym and practice, and your test scores go up. Great for you. They’re showing test score growth, but student selection is probably also shaping their data set. And it’s really optimizing for the old testing system.
The most positive view I have of Alpha School is that they’re basically saying, “These metrics are a total waste of time. Let’s just jam them into this two-hour window, and then the rest of the day, the kids can actually learn meaningful things—developing real competencies, doing big projects, all of that.” I’m far more interested in what’s happening the rest of the day than in those two hours.
I hope personalized learning starts to look more like project-based learning, where AI provides scaffolding (or not) for a team of students, or even a whole class, as they move through meaningful benchmarks. In that sense, it looks a lot more like my business school experience: we did simulations, case studies, and projects, almost always in teams of two to six. That’s what the majority of K-12 education should probably look like.
Tutoring is also getting a lot of attention right now—can we get to two-sigma impact? But most people don’t understand that tutoring is not a search bar. Tutoring is an experience in which an adult guides your learning by understanding your competency level and moving you through the exercises that help you achieve mastery. Any attempt to personalize tutoring with AI has to take that as a first premise.
I think the best models use near-peer tutors—a college student tutoring a high schooler, say—using AI to level up the quality of the tutor and provide the data analytics to better understand what’s in the student’s zone of proximal development.
Hopefully we don’t repeat those mistakes; unfortunately, I do think some of these misunderstandings are persistent. We can’t be content putting kids into “personalized” learning experiences, sitting alone with headphones on, drilling for test scores. No parent wants that for their kids. We need to stop building a system we wouldn’t want our own children to participate in. Instead, let’s build one that helps them thrive in a much more dynamic and collaborative future.
ALLISON: Let’s talk a little bit about what that would look like on a high school level. What would a from-scratch AI native high school look like?
BEN: I think that would actually look like a portfolio of schools, and some of those models already exist and could be elevated with AI. One is the job-embedded high school, where students get work experience, credit, and build competencies—kids are job-placed as they get older, and as they progress, the ratio goes up from two to three days a week on assignment. They’re still getting instruction and support too. Bigger Picture Learning has a network of these schools.
The classic magnet school is another model: essentially, these schools bring together kids across interest areas—STEM, the arts—and let them do meaningful projects in those domains. One of the biggest drivers of performance in high school is your peer community; bringing together like-minded peers creates the opportunity for that kind of engagement and outstanding, spiky work.
In the 1990s, when charter schools were growing but not yet the threat to the public system they’ve become, superintendents were thinking a lot about developing a portfolio of six or seven high schools—maybe two or three schools, with several different programs—that were interest-based or competency-based. They wanted to give families more opportunity and choice to pursue what’s meaningful for their students. Writ large, that’s the right future for high school. And it’s a great time to be in school design, because there’s a lot of room to be creative and locally based while still bringing rigor and structure to the model.
On The Unbundling of the School-Based Learning Market
ALLISON: Let’s talk a little bit about education savings accounts (ESAs) and the unbundling of the learning market that comes along with them. Within the vision you’re outlining, what do you see as the big opportunities for ESAs to participate, and what risks would you call out from the start?
BEN: In general, I’m very excited about what ESAs will mean for innovation in the sector and for ed tech pathways. But I think the process will be uneven unless we develop a meaningful quality signal and control.
Like any industry going through an unbundling of payment, education will experience a sea change. When you transition from the government both paying for and providing the service, to the government paying for a network of independent providers to deliver it, you can see a huge impact on outcomes. Healthcare is a good corollary: that industry moved from HMO delivery profiles to a range of payment and treatment options, and the net impact has been really good. Decade over decade, health outcomes have improved.
But in an unbundled market, if access is high, so is variability in quality. Therein lies the challenge of ESAs—there’s no quality accountability in any of these systems. Families can now use those dollars in many different ways: some will go all in on homeschooling, others will subsidize private school tuition, some will take a hybrid route. But the outcomes are nobody’s job to guarantee.
I’d ask every state superintendent this: “Do you have an assessment system you can incentivize across your ESA portfolio that provides diagnostic and benchmark data to understand which strategies work for which populations?” That transparency is necessary for families to make the most of this new market and choose the right services for their kids. It also helps districts see what’s winning out.
We aren’t yet wrapping our heads around the consequences of this unbundling. Look at the polls: across every group—Democrats, Republicans, African-Americans, Latinos, Whites, Asians, poor, middle class, and very wealthy—80% support education savings accounts. Americans agree public education is broken, and we’re collectively desperate for other options.
ALLISON: It’s long been said that innovation happens either in the unbundling of products and services or in the rebundling of them. And rarely have we had the window of opportunity to think about what it actually looks like to unbundle the public school system.
BEN: Right. Just as traditional school funding recedes, unbundled funding is reshaping the marketplace. That’s a new field of opportunity in K-12 ed tech, and I think legacy companies may have a hard time pivoting to this opportunity. AI-native companies simply need to convince a family their product is best for their child—and parents don’t even have to pay for it, since the ESA does. That’s where all the action will be over the next five years.
On Why Philanthropy Should Focus on Assessment
ALLISON: Philanthropy occupies this rare ability to move in a way that’s aligned with public good, faster than public institutions. If a major philanthropist gave you 200 million dollars or more for that matter, to execute the vision that you’ve been outlining, what would you fund?
BEN: I’m excited to answer that. First, it’s important to caveat the philanthropic landscape of education. The last generation of tech billionaires was burned by it: Mark Zuckerberg and the Newark debacle; Eric Schmidt, whose Schmidt Futures has entirely dismantled its education function; and Bill Gates, who has twice publicly admitted his foundation’s strategy over the last decade has failed. Marc Benioff gave $30M to Oakland Public Schools, and six months later they were in deficit. New donors are actively being steered away from education, because there’s a profound distrust in institutions’ ability to steward donations. That includes higher ed: donors and the public are increasingly skeptical about how well these institutions handle the money they’re given.
We in education can’t squander this moment as new millionaires enter philanthropy, because philanthropy fills in what the marketplace won’t do for itself.
If I had $200 million to invest in building the learning system for the 22nd century, I’d radically redesign assessment. Working with the best psychometricians and state education chiefs, I’d develop a system of summative diagnostic testing of both individual and collective performance. From that spine would emerge innovations across pedagogy, curricular tools, products and services, and delivery modes—essentially a decentralized innovation layer, because everyone is now running from the same scorecard.
That’s what’s been missing. When the New York Times reports on a decade and a half of declining test scores, it doesn’t mention that we lost our accountability system at the same time. No Child Left Behind didn’t codify the right measures, but it was the right strategy: to create a language around what student performance should look like and what support we should have, for life. Learning, like health, is a continuum—not black and white, not “proficient” or “not proficient.” We need a way to measure that development over time. Because ultimately we now need a nation of learners, from kindergarten to the grave.
One Small Signal
ALLISON: What’s one small signal in the world right now to which we should be paying more attention?
BEN: Kids are ready to build. The entrepreneurship that I’m seeing from the 10-14 year old set inspires me. Kids don’t need to wait for someone else to solve the problems they experience—they can build the solutions. Everyone will be empowered to solve their own problems by building. I wish we could see the four years of high school as the zone of innovation, entrepreneurship, and economic mobility, because kids are ready for it.
Ben's line about kids sitting alone with headphones on, drilling for test scores, gave me a pause — because that's exactly the image that comes to mind when most people hear "AI-personalized learning." And I think that image is doing real damage to the conversation.
The ceiling of AI-assisted learning isn't efficiency. At its best, it's the possibility of genuinely knowing a child — their curiosity, their pace, their gaps, their strengths — and using that knowledge to put better learning in front of them, so the human around them can do more of what only humans can do: build the relationship, hold the culture, notice the child who's struggling for reasons beyond the curriculum.
In that sense, AI doesn't replace the human in education. It gives the human back. In homeschool context: a parent who isn't spending hours designing curriculum or tracking gaps has more capacity to take a walk, follow an unexpected interest, or simply be present. A teacher freed from pacing guides and compliance workflows can notice the child who needs to be pushed, or the one who needs to slow down.
I'm new to this space and still forming my questions. But the one I keep returning to isn't what AI can teach a child — it's what it makes possible between a child and the adults around them.