Beyond the Pilot: What Districtwide AI Integration Actually Looks Like in Year Two
Year one of AI in schools runs on adrenaline. A few enthusiastic teachers, a splashy pilot, and a board presentation that lands well. Year two is where you find out whether any of it was real. The novelty wears off, the hard questions arrive, and the honest superintendent stops asking, “Are people excited?” and starts asking, “Is this actually changing instruction, and for whom?”
This June, I retired after 16 years as an Illinois superintendent, the last eight leading the nine-school PK–8 district whose Innovation with Guardrails framework anchored the first post in this series. That timing matters. What follows is not a progress report I have to defend at a board meeting. It is what the second year’s data said as I handed the work off, including the parts I’d rather it didn’t.
The Adoption Is Real, and It’s Sticky
The first thing year-two data settles is whether the pilot was a peak or a plateau. Ours held. The 74,000 staff AI generations I cited in the first post matter less than their shape: more than 3,000 a month in 15 of 18 months, with the dips tracking summer break, not disengagement. More than half were everyday teaching work, including drafting, rewriting, and feedback. That is not a demo anymore. It is infrastructure. And infrastructure, unlike enthusiasm, does not leave when the leader does.
The “PD Fatigue” Story Is More Complicated Than I Expected
I went looking for burnout and found an appetite. In our staff survey, 40 percent said AI was underutilized in their schools. Only 12 percent said it was overutilized. Fifty-six percent asked for dedicated time to collaborate with colleagues on AI-supported instruction.
The fatigue is not with AI. It is with learning it alone. That is a professional learning design problem, not a motivation problem, and it is the clearest mandate the data handed the district for year three.
More Familiarity Produced More Skepticism, Not Less
The teachers most fluent with AI were also the ones raising the sharpest concerns about training-data quality, bias, student data privacy, and the long-term effect of scaffolding on foundational skills.
The pattern repeated with students. Our sixth graders posted the largest skill-confidence gains in the entire study while their overall optimism about AI held exactly flat. They got more capable without getting more credulous.
The student line I closed the first post with, AI is “a tool, not a friend,” was not a slogan we taught. It is the disposition the data kept finding. Capable skeptics are exactly what we should want, and structured exposure is what produces them.
Is It Changing Instruction? In Some Places, and We Can Prove It in Some Cases, but Not All
Here is where I will be honest about the gaps. We built rich evidence inside our four structured cohorts: students building functional apps, revising persuasive essays across two dozen iterations, and defending claims against an AI debate partner.
But beyond those cohorts, our platform dashboard logged more than 16,000 student engagements across 337 teacher-created AI spaces. We never broke that figure down by building, which means we could not yet say whether the benefits reached our two very different communities equally.
The data exists. When I left, we had not yet asked it that question. Naming that gap publicly is the precondition for closing it, and it sits at the top of the year-three docket I handed off.
The Policy Gaps You Find Reveal the Policy You Needed
Year two surfaced friction that year one never could: internet filters blocking the research sites students needed, open-ended tools that worked better with explicit boundaries, and an AI feedback tool that scored generously enough to caution against treating its numbers as grades.
None of these were failures. They were the system telling us where the next version of the policy had to go. A pilot cannot surface these issues. Only scale can.
So what is year two, really? It is the year the work stops being about excitement and becomes about evidence, equity, and design.
And here is the part I could not have written a year ago: Year two is also when you learn what the work was attached to. A binding board policy, a tiered assignment framework, students with voting seats, and a community that helped draw the lines were never meant to be my initiative. They were meant to be the district’s operating system.
Guardrails that depend on the leader are not guardrails. They are personality.
In July, I watched my successor take the wheel of a system with durable adoption, discerning users, and a clear, uncomfortable list of what it still owes its community. Year two tells you whether the work is real. The transition tells you whether it was ever about you. The best possible answer is no.
