A plug before I begin. This essay argues that the teachers who can carry AI work are already in those buildings, and that the field’s job is to find and back them. Spark the Future is the aiEDU program where we do that. Applications for the first national cohort close soon — about ten seats left. If that’s you or someone you know, here’s the link. Now to it.
+++
Every new general-purpose technology arrives the same way: ahead of the institutions that have to absorb it. The hard part is rarely the technology itself. It’s the change management — the slow, uneven work of getting humans to redesign how they work around a new tool. Personal computing took roughly fifteen years to deliver the productivity gains everyone kept predicting — not because the technology lagged, but because organizations had to learn to redesign work around it.
AI is in that stretch now, and K-12 is the sector most exposed.
The default district response is to treat AI like every prior technology adoption: vet the tool, write the policy, train the staff, publish the guidance. That work has to happen. It also isn’t the bottleneck. The bottleneck is human capacity to take on a new way of working, and in a school district, that capacity gets built or broken classroom by classroom.
I came into this work without a background in education or technology. For a long time I wasn’t sure what I was adding to a field crowded with people who had both. The instincts my team and I keep returning to come from somewhere else: I started my career as a community organizer. What organizing teaches you is that information doesn’t move through a closed institution on its own. People do — the right people, when the conditions for them are in place. Most of what’s happening in AI-for-K-12 right now is the opposite. Vendors, agencies, foundations, and consultants, standing outside the building, broadcasting in.
Schools have been here before
K-12 has a long history of new technology arriving with big promises and producing almost no change in practice.
Larry Cuban, the Stanford historian who’s spent forty years studying this, has a useful term for why, which he calls the grammar of schooling, that is, the aspects of schooling that stay constant across locations and generations. From the familiar bell schedule and the age cohorts to the period blocks. That language is older than any of the tools we keep introducing into it, and it bends them all to fit. Personal computers in the 80s, Smartboards in the 2000s, iPads in the 2010s all came with promises of democratization and personalization. And yet none of them changed how teachers teach or how students learn in any deep way.
Schools absorb new tools the way they do because of how they’re structured. They’re closed systems—institutions where the people on the outside, no matter how well-resourced or well-intentioned, can’t actually reach the people on the inside without help. Procurement decisions get made several layers above the classroom. Teachers often find out about new tools the same week they’re expected to use them. Professional development is episodic, external, and disconnected from the actual work of teaching. A teacher’s day is spent alone in a room with twenty-five kids, with almost no time built in to learn anything new.
From the outside, schools are pawns in the grand strategy of tech transformation, with vendors, advisors, and departments of education working throughout the field. From inside the building, though, very little of it gets through.
Closed systems aren’t unique to K-12. Hospitals are closed in similar ways. So is the military. So is most of corporate America once you get past the front door. What they have in common is that information and new practice don’t move through them the way they do in open systems. They move person to person, through trust networks, only when someone inside decides to carry them.
Community organizing is the practice of moving information through those trust networks deliberately. I learned it in a context that had nothing to do with AI but everything to do with how change actually spreads.
What I learned organizing in Ohio
I started my career on the Obama campaign, running on-campus operations at Ohio State. OSU is in Franklin County, then one of the most consequential counties in the presidential election cycle. The campus had close to 50,000 students, most of them in dorms. Student residence halls had restricted keycard access, posing a challenge for voter registration sweeps which involve painstakingly knocking hundreds (for some volunteers, thousands) of doors.
I had a small budget and a hard problem: how do you run voter contact inside a closed environment when you literally can’t get through the door?
Other people have to do it for you.
I built a recruitment funnel for students who already had access to the dorms I couldn’t reach. I pitched the work as an internship instead of political volunteering, solving a problem for students who had not yet lined something up for the summer. They needed experience to add to their resumes, not yard signs. This required some upfront work to build some additional structure, weekly check-ins, and leadership ladders to model internship programs. This wasn’t standard for a field campaign at the time, but all of it turned out to be a necessary component of empowering the students to truly own the work themselves.
By election day, my interns had spread across campus at a density no paid staffer could have matched. Turnout went up about eight points over the previous cycle — 2008, already itself a landmark year for youth turnout.
What I learned that year is something almost every technology rollout I’ve watched since gets wrong. You can broadcast a message at any volume. You can buy the best media. You can publish the best framework. None of it gets through a closed door. What gets through is one person inside who already has the trust of the people you’re trying to reach, and who decides the work is theirs.
Preparing every student for AI is the campaign I’ve dedicated my life to. Schools are the closed dorm buildings. And most of what’s happening in AI-for-schools right now is people standing outside, trying to broadcast in.
The wrong question
This sounds obvious. It isn’t what’s happening today. The “AI + education” conversation is dominated by helping schools answer one question: what should we buy, and what should we tell teachers to do with it?
The venture-backed ecosystem is happy to answer that question. And they have hundreds of millions in funding to help them do it with legions of marketers and sales teams to flood the zone. MagicSchool, SchoolAI, Brisk, and Cluely alone have raised more than $130 million in venture capital over the last three years. To be fair, three of these are legitimate edtech companies; Cluely is the one unambiguous bad actor list, with a marketing campaign that literally promises to help students “cheat on everything.”
Those four are just a slice of the picture. The figure doesn’t include OpenAI, Anthropic, or the other frontier-model companies. It’s not obvious how you’d disaggregate an education-specific dollar figure from the rest of their business, but given OpenAI and Anthropic together have raised more than $250 billion, even a fraction of a percentage point puts the answer in the billions.
Meanwhile, if you combine the annual budgets of some of the most prominent organizations doing the ground- and systems-level organizing—Digital Promise, InnovateEDU, Code.org, Quill, ISTE, Playlab, Leading Educators, CSTA, AI for Equity, Raspberry Pi Foundation, CRPE, and Learner Studio (to name just a few!)—you’ll end up with a number that is still smaller than the compensation package being offered to Matt Deitke, a 24-year old AI researcher who accepted a $250 million offer to join Meta’s Superintelligence Lab last year.
It’s best not to get caught up in the hyper-capitalism of the AI age. The numbers are just silly.
In any case, it isn’t a question of edtech OR bottom-up change management. The companies vying to sell technology into schools won’t succeed without it.
What teachers and students (and edtech companies) actually need is the human capacity to direct AI, build with it, evaluate what it produces, and know when to ignore it. That isn’t something a district can purchase. It’s a practice that has to be developed in context, alongside other practitioners, over time. It gets harder to build, not easier, the faster the technology moves.
From Polson to Akron
When I think about what this looks like when it’s working, two stories stay with me — Connor Mulvaney and the state of Ohio. Connor is what one teacher can do when the conditions are right. Ohio is what a statewide network of teachers builds over years. In both cases, we got there by finding the teachers willing to lead this work, developing them into the people who could, and staying long enough for the network around them to form.
Connor was a science teacher in Polson, Montana (a rural community in the northwest corner of the state) when he came into our Trailblazer Fellowship as a member of our inaugural cohort in 2023.
He built AI literacy with his own students by asking three questions: What is AI good at? What is AI not good at? What’s one thing you want to learn about AI? He coached colleagues across his district. He partnered with CAST and the Boys & Girls Clubs of the Flathead Reservation and Lake County to launch AI in the Big Sky, a Montana-specific initiative for rural and Indigenous students. He now leads our Trailblazer cohorts for 130 educators nationally.
I’ve learned a lot from Connor. The teacher we trained is now one of the people I learn from. That’s something you don’t get from just dropping a curriculum.
It started in 2019 at Firestone High School in Akron: a few teachers willing to try something new with their juniors and seniors, an administration willing to let them, and within months a small grant from Macy’s to fund the pilot. There was no curriculum yet, and no organization to speak of. By 2021, the work had grown into a seat on the state’s Computer Science Standards Revision Advisory Group. By February 2024, the Ohio AI in K-12 Toolkit launched, published by the Lieutenant Governor’s office. Across 2024 and 2025, eleven regional summits reached thousands of educators, co-hosted with Ohio’s Educational Service Centers, along with monthly meetings with dozens of administrators across the state.
In 2025, Ohio HB 96 passed. It’s the first state law in the country to require an AI policy in every public school. Roughly six hundred Ohio districts have until July 1, 2026 to comply. The model policy they’re working from is one we helped write.
Six years from a few teachers at Firestone to a state mandate. None of that happened because we pivoted into policy. It happened because for six years we kept showing up in schools and building . The summits, the policy mandate — those weren’t the strategy, they were what the strategy let us do, the byproduct of having spent six years building a network strong enough to carry them.
Two failure modes
I want to be careful here, because the model I’m describing has two failure modes I think about a lot.
The first is that teacher-led organizing slides into unpaid teacher labor if you aren’t careful about it. RAND found that K-12 teachers report about 53 work hours a week, with 15 of those uncontracted and 12 uncompensated. Pew found that 84% say they don’t have enough regular work hours for what they’re already responsible for. Those numbers describe my mother. She retires from Akron Public Schools next month. She stays late. She tutors students after the bell. She teaches at the University of Akron in the evenings. She shows up to school plays and dances on weekends. Sixty hours a week is conservative.
If our model for AI in schools depends on more of that, the only teachers who can sustain it will be the ones with spare time and supportive principals, and the base will narrow instead of widening. So paying teachers properly for this work has to be a design principle, not an aspiration. Our Trailblazer Fellows are stipended. Cohort leaders are paid. We push districts to put teacher-leader roles inside the workday, not after it. Teacher-led can’t mean teacher-loaded.
The second failure mode is the opposite, and it’s the one I think about more. “Teacher-led” can quietly turn into leadership stepping back. Getting AI right in a building still rests with the people who set the conditions for everyone else’s work: superintendents, principals, department heads. The shift isn’t whether they lead. It’s how. Their job is to make the conditions inside their buildings hospitable to the people already doing the work. Paid time. Political cover. The authority to step into a leadership role peers will follow. That’s still leadership. It’s harder leadership, in a lot of ways, than the top-down version.
I hear something similar from the CEOs I talk with. Most CEOs frame AI to their workforce as a productivity play. That reliably backfires. Productivity is valuable to a CEO in a way it usually isn’t valuable to the person doing the actual work, and a workforce that hears “productivity initiative” hears “we’re trying to do more with fewer of you.” The companies seeing genuine buy-in have figured out how to frame AI instead as a chance to give their people transferable skills — skills they’ll carry through the rest of their careers, with us or somewhere else. That framing only holds up when leadership and the ground share it.
We’re in the window now
There’s a version of the next ten years where the schools that figure this out get a generational head start, and the schools that don’t get further behind. The teachers who can lead this work are already in those buildings. They aren’t waiting for permission, and they aren’t waiting for a vendor to finish vetting. They’re already doing it, in pockets, in places like Polson and Akron, often without the time, structure, or recognition the work deserves.
In August 2023, I told WorkingNation that we don’t have twenty years to get AI in education right. We have three. We’re now inside that window.
The job isn’t to invent the people who can carry this work. It’s to find them, back them properly, and figure out how to remove the barriers that are stifling the desperately needed cultures of innovation demanded by school transformation in the age of AI.