From Dan Fitzpatrick, The AI Educator
Independent thinking on AI leadership, strategy, readiness and organisational change. Grounded in research and the realities of working with leaders.
Explore the key questions
The leadership agenda
Latest thinking
60 insights
Treat an AI error as misconduct and you will not hear about the next one. Treat it as nothing and the rule was decorative. The Rule-First Test asks whether the organization had a rule specific enough to break before it asks anything about the person.
Most school AI roadmaps fail on sequence, not dates. A roadmap should name five gates, each a condition that has to be true before the next phase can deliver anything, with an owner, evidence and a check date.
You cannot prove AI caused an outcome in your school, and promising a board you will is how leaders end up defending figures they cannot source. The Evidence Chain is four links, and the honest report names the one that breaks.
Most schools that ask what belongs in an AI policy already have one, and it still cannot answer the question a teacher asks on Monday. A policy is a record of decisions, not a document you write. Here are the eight decisions it has to record.
Most AI training produces confidence, not capability. Three questions separate training that changes what your staff can actually do from training that leaves them dependent on the tool.
Commercial AI detectors are too unreliable to start a disciplinary process. The decision that matters is not which detector a school uses, but what a high score is allowed to trigger, and who decides.
Ofsted sets no AI standard and requires no AI policy. What it may ask is how you make sure your use of AI supports the best interests of children. Here is what that question is really asking, and the four-part answer a leader should be able to give without notes.
KCSIE 2026 changed one definition, not your whole safeguarding duty. Here is what the statutory guidance actually says about AI, where the real requirements sit, and the four checks worth doing this term.
When a member of staff asks whether AI will take their job, they are asking about their tasks, not their title, and most of those tasks have already moved. Four answers a leader can give that are true, and the one promise you can actually keep.
Leaders who feel outpaced by their staff on AI try to close the gap by learning the tool. That is the wrong gap. Operating skill spreads on its own; understanding what the organization has already changed does not, and only a leader can supply it.
Most AI items on a board agenda inform governors without ever letting them govern. Here is a four-line board paper, a test to run over it before you send it, and three things boards are currently being told that the source documents do not support.
Ask twelve heads in one trust who approves a new AI tool and you get four answers. The trust owns the decisions that must be the same everywhere; each school owns the rest. Here is the test that sorts them, and the four a trust cannot delegate.
