Four things a fluent leader can do
Stated as capabilities rather than topics, because a board can test a capability and cannot test a topic.
Judge the output without re-doing the work. Knowing your own domain’s failure patterns well enough to see where a plausible answer went wrong. Confidence is a property of writing style rather than of knowledge, and this is the single most valuable thing a senior person can still do that a machine cannot do for them. How do I know when AI is wrong.
Know which decisions to keep. Not everything that can be delegated should be, and the test is not difficulty. What stays human, and which decisions should become slower.
Use it without losing the thing you are paid for. Keeping the repetitions that build the capability, delegating the ones that do not, and noticing the difference before it is gone. Using AI without dependency, and capability debt.
Answer for it. Being able to say what the machine contributed, what you decided, and on what basis, to a regulator, a journalist, a board or a court. Proving you did the work.
Why this is not a training course
Training builds a skill against a curriculum and there are firms that do it well. If what you want is a workflow built and a hundred people taught to use it, use one of them and you will get a better result than you would here.
This is the part that does not transfer from a curriculum: judgement in your own domain, applied to your own decisions, with somebody beside you who will tell you when the answer you liked was the wrong one. It is closer to advice than to teaching, so it is sold by the session rather than by the seat.
Three ways it runs
For a board. Directors are in an odd position: accountable for the oversight of systems they do not use and often cannot see. The board version is about reading what management brings, knowing which questions have answers, and the one nobody asks out loud, which is whether directors should be putting board papers into an assistant. That question has an answer, and not the obvious one.
For a chief executive, one to one. Private, in a peer register, on the decisions AI arrives at without answers to. What to hand over, what to keep by hand, and how to tell which is which when the tool makes both feel identical.
For a senior team. Half a day with the executive, because the decisions a board wants made are usually made a layer below it, and a team that has not agreed where its own line sits will draw it differently in every function.
The senior team version runs as an AI thinking workshop rather than a fluency one, and the word is deliberate: how to prompt AI for strategic decisions, how to challenge AI outputs rather than accept them, and how to build habits of critical judgement alongside AI use. For senior leaders and team leads, not technical staff.
The obligation most boards have not noticed
Article 4 of the EU AI Act has required providers and deployers to take measures to ensure a sufficient level of AI literacy among their staff since 2 February 2025. It applies to a UK company that places AI systems on the EU market or whose system output is used in the EU, it is already in force rather than pending, one of the few AI obligations that is about people rather than systems.
It is worth being precise about what that does and does not mean, because it is being oversold. It does not prescribe a syllabus, a certificate or a supplier. It asks an organisation to take measures appropriate to the role and the risk, and a board that can describe what it did and why is in a better position than one holding a completion certificate. What AI literacy means for leaders.
What the evidence says about losing the skill
This is the part that makes fluency a governance question rather than a development one. Skill that is not used degrades, and it degrades faster and more than most organisations expect. The research’s evidence base grades every study behind that claim, including what each one does not show: is deskilling real, how fast skills decay, and whether you can regain a skill you have lost.
A leadership team that has never asked which of its own capabilities it is spending has usually spent some.
