A year ago, school leaders asked us how to get their people comfortable with AI.
This fall, they are asking harder questions: How do we keep students safe? How do we measure impact? Who owns the decisions? And what should we scale?
AI is already entering schools. Leaders now have to shape how it does.
This is an important shift: Systems are moving from early-stage adoption to institutionalizing accountability. The curiosity and experimentation of the first wave have been joined by a much clearer understanding of the risks, unknowns and responsibilities involved.
We see it across ILO Group’s AI Strategy and Leadership Network. In two school years, 68 systems in 28 states have participated, including 15 state agencies and more than 215 leaders collectively serving 15 million students.
The systems joining the network this fall are not the same ones that participated last year, so this is not a year-over-year scorecard. But what leaders arrive with, and what they are asking for help with, paints a useful picture of where the field is now.
Last year, two out of three participating districts told us that student AI use was largely up to individual teachers. This year, half of the systems in our latest cohort have common rules for student use by grade or school level.
Last year, most districts arrived having already trained their staff, and most had a cross-department AI team. This year, half arrive with that team, 1 in 5 with cybersecurity training for staff, and almost none with training built for the specific jobs people hold in schools.
There is change at the state level, too. Last year, the dominant work was publishing guidance. This year, 3 out of 4 participating state agencies provide statewide training and implementation support. For the first time in the network, one state is funding competitive pilots and another is providing research support.
And for the first time, participating leaders are asking us a question that ought to define the next phase of this work: How do we know whether any of it is actually working?
Getting to that answer requires asking a few more questions. Four of them.
1. Is this tool safe, and is it actually good enough for students?
It’s one thing to meet technical requirements; it’s another to actually improve instruction. An AI tutor can pass security reviews and still not produce learning gains. A tool can function exactly as designed and still give students answers too quickly, undermine productive struggle, or fail to support the instructional approach in the classroom.
So leaders need to ask more than whether a tool can be deployed. They need to know whether doing so benefits students. The goal is to build a disciplined way to make good decisions despite them.
Delaware is building that capacity at the state level. Under the leadership of Secretary Cindy Marten, the state is developing an AI assurance lab that tests tools, gathers teacher feedback, surfaces early concerns and helps districts evaluate products so procurement rests on state-validated evidence rather than vendor promises.
Dallas ISD has built a district-level gate. Every AI resource added to the district’s systems goes through a vetting process by its technology and academic department and is reviewed for safety, privacy, educational value and alignment with district goals before staff or students can use it. The department reviews tools again regularly after deployment.
Not every district can build an assurance lab, but every district can build an assurance process and define what it expects a tool to do educationally and from a safety perspective. It can require evidence that tools meet that standard before putting them in schools, and continue to test them afterward with clear and transparent criteria.
2. How will we know whether it works?
The largest group of districts arriving this fall is still running small pilots. But leaders are no longer satisfied simply to say they piloted something. They want to know whether it produced a result worth continuing.
That requires discipline. Every pilot should begin with four things: a named owner, a success measure established before launch, a decision date and a required recommendation at the end to scale, revise or stop.
“Revise” is a legitimate outcome. Refining applications of AI builds organizational capacity. What should disappear is the perpetual pilot: an initiative that rolls from semester to semester because “success” was never defined and a decide-by date wasn’t established on day one.
The critical metric will depend on the use case: administrative time saved for a teacher, consistent implementation of policy, strengthened instructional quality, or demonstrated learning improvements. For an operations-focused tool or application, it may be accuracy, reliability or cost.
The important thing is that it is determined at the outset and measured.
3. Who is responsible for making the decisions?
This year, more participating districts report having purchased enterprise AI licenses than having a board policy, an approved list of tools or a dedicated budget. That sequence is understandable. The technology moved quickly. School systems responded.
But licenses do not create strategy. And in the absence of strategy, rather than stopping, adoption simply happens without a coherent system around it.
Most systems cannot afford a chief AI officer, and most do not need one. What every system does need is a shared understanding of who can approve a use case, who assesses risk, who reviews instructional value and who has authority to say yes, no or not yet.
Governance should match the organization. A large, technically complex system may need dedicated leadership. Another district may need a senior leader with protected time and a cross-functional team. What matters is less the title than the clarity: defined decision rights, common risk levels and one accountable owner for each initiative.
The goal is not more bureaucracy. It is to build the systems and guardrails that let leaders make decisions quickly enough to keep pace with the technology without sacrificing judgment.
4. Are we building capacity that will last?
Systems must build the ability of adults to use AI well.
One-time and opt-in training remains the most common model among districts arriving this fall. That kind of training can introduce a technology, but it does not build lasting capability or embed enduring competency.
Ector County ISD in Texas, under Superintendent Keeley Boyer, spent a full year building foundational AI understanding in principals, the school board and across the leadership team before expanding the work into classrooms with training for teachers by grade band.
That approach takes longer than offering one session to everyone. But principals, teachers and central-office leaders make different decisions and face different risks. Their preparation should reflect that. As the technology keeps changing, the work still has to come back to the question educators have always asked: What will help students learn and thrive?
The same principle applies beyond individual districts. Capital Region BOCES in New York is building its own organizational approach to AI while helping the districts it serves develop guardrails and guidance, exchange what they are learning and strengthen career pathways with the operational support behind them.
States can play the same role at a much larger scale: creating ways for districts to share which AI products and systems they are using, what they tested, what worked and what they stopped doing. States also play a critical role in contract negotiations, standardizing expectations of vendors, overseeing data agreements and supporting outcome-based contracts.
But states also need capacity to sustain the work. In at least one participating agency, a single person is responsible for AI both inside the agency and across the field. Most participating states are funding the work with discretionary dollars.
That is difficult to sustain. AI cannot simply become one more responsibility added to an already full job. Systems will eventually have to decide not only what they are starting, but what they are willing to stop.
THE KEY QUESTIONS WILL CONTINUE TO CHANGE
A year ago, the challenge was to get people to effectively engage with AI. This year it’s about understanding if that engagement is productive.
Next year the burning question will change again. No system will be able to fully predict or control what comes next.
The leaders who respond strategically will be the ones who built governance this year that creates the capacity to test, measure, decide and adapt as the technology changes.
That is both the infrastructure states and districts need right now and the lasting competency this next era will require.
Julia Rafal-Baer is the co-founder and CEO of ILO Group, a women-owned education and policy strategy firm, and the founder of Women Leading Ed, a national nonprofit network for women in education leadership. She is also a member of the National Assessment Governing Board.
Scott Muri served as a superintendent for 10 years at two large Texas school districts (Spring Branch ISD and Ector County ISD) and is now the CEO of innovations in leadership and superintendent in residence at the policy firm ILO Group.
Leave a Reply