Twenty-five years in education data, and the hardest problems have almost never been technical. They are definitions nobody agreed on, ownership nobody claimed, and one number that three offices each define differently. AI does not fix any of that. It makes it louder.
An AI Operator does more than produce prompts. An operator frames the work, supplies context, establishes boundaries, verifies results, and remains accountable for what happens next. Four disciplines carry that responsibility.
Define the problem, audience, purpose, decision, and quality standard before choosing the tool.
Give authoritative sources control over fluent inference and make unsupported gaps visible.
Match review to consequence and use a qualified person who can recognize a plausible error.
Define ownership, information boundaries, monitoring, correction, and the authority to stop.
Enterprise capability, sized for agencies that count every dollar and often have one person doing the work of five.
Most of the AI conversation in education is about students. The larger opportunity sits with the adults who administer programs, answer families, reconcile reports, and interpret policy from information scattered across a dozen systems.
Explore AI in Education →A policy can describe where AI may be used and who answers for it. It cannot make those decisions for you. Governance is what connects acceptable use to real work, named owners, approved sources, and a stop mechanism that is more than a reassuring sentence in a document.
Explore AI Governance →Agencies rarely lack data. They struggle because definitions conflict, ownership is unclear, and reasonable people cannot agree on which source to trust. Buying a catalog will not settle that. A catalog is furniture.
Explore Data Governance →This site holds the argument in short form. The full pieces are on Substack.
A capable AI can still invent citations, infer unsupported numbers, and repeat outdated guidance when education professionals give it no context, standards, or boundaries. The answer is disciplined onboarding: define the work, supply trusted sources, explain what failure costs, and keep a qualified person responsible for verification.
Read the article on Substack →Sessions for education agencies, conferences, and leadership teams, built on twenty-five years of doing the work rather than watching it.
The part of the job that does not transfer to the tool.
What has to be true before an agency relies on an answer.
Why buying another tool has never once fixed it.
Every session is built around one question: what decision gets better because we did this? It keeps a talk away from impressive demonstrations and on the work an agency still has to do on Monday morning.
Ten years as a full-time youth minister. Then my father handed down his old IBM PS/2, I spent six months learning OS/2 for no practical reason at all, and in 1995 a friend called the church office asking whether I knew anyone with OS/2 experience. I did.
Everything since has been the same job wearing different clothes: making information useful to the people trying to serve children.
About Mike →Speaking invitations, professional questions, and media inquiries are all welcome. If you are stuck on a governance problem and want a second opinion, that counts too.