Schools are quietly figuring out how to use AI without breaking education

Teachers aren't waiting for perfect policy. They're testing AI in real classrooms right now—and some experiments are actually working.

The conversation around AI in schools usually splits two ways: either it's going to revolutionize learning or destroy it. But at Harvard's Graduate School of Education, a different group is asking a more practical question: what does AI actually do well in a classroom, and how do we build it with teachers, not around them.

The answer looks less like a Silicon Valley rollout and more like careful, local experimentation.

Tailoring Learning to the Student

Yenda Prado, a research analyst at Digital Promise (an education nonprofit), is working on something called "learner profiles." The idea is straightforward: AI systems trained on data about how different students learn—English language learners, students with dyslexia, kids who thrive with visual instruction—can adapt lessons in real time. Instead of a one-size-fits-all math worksheet, an AI tutor adjusts its approach based on what actually works for that specific student.

This isn't about replacing teachers. It's about giving them better information. A teacher with 30 students can't easily track exactly where each one gets stuck. An AI system can. The question Prado's team is asking is: how do we build that feedback loop so teachers actually use it?

Stripping Math Down to What Matters

Kedaar Sridhar, a 2025 Harvard Education Entrepreneurship Fellow, took a different angle. He built an AI platform that analyzes math curricula and asks a simple question: what's actually essential here, and what's just clutter. His platform at M7E AI focuses on what researchers call "productive struggle"—the kind of challenge that builds real understanding, not the kind that just frustrates students.

The insight is almost obvious once you hear it: most textbooks are bloated. They repeat concepts, layer in unnecessary variations, and make it hard for students to see the core idea. An AI system can strip that down, leaving room for students to actually think.

Testing at a Smaller Scale

Keith Parker, superintendent of Elizabeth City-Pasquotank Public Schools in North Carolina, took a different approach. He built a microschool—just 25 students, 3 teachers—as a testing ground for AI integration. At that scale, it's possible to try things. Use AI as a supplemental tutor for some students. Use it as a primary instructor for specific units. Watch what happens. Adjust.

Microschools aren't a solution for every district. But they're useful for answering the question most schools actually care about: "What does this look like when we try it?"

The Real Shift: Involving Teachers

What ties these experiments together is something less visible than the technology itself. All three approaches involve educators from the beginning. Prado emphasizes that researchers and developers need to work directly with school communities, testing ideas with real teachers, iterating based on feedback, and being honest about what AI can and can't do.

This matters because the gap between "AI can theoretically do X" and "teachers will actually use AI to do X" is enormous. A system that sounds great in a lab might feel like extra work in a classroom. A feature that works for one group of students might confuse another. The only way to know is to ask the people who spend eight hours a day with students.

The educators involved in these projects agree on one thing: we're in the middle of a significant shift in how K-12 education works. But the shift won't happen because of AI itself. It'll happen because schools are choosing to be intentional about it—testing specific solutions, staying grounded in what students actually need, and keeping teachers at the center of the decision.

That's not revolutionary language. But it might be the only approach that actually sticks.