AI fluency for leaders starts with the right question
Walk into almost any leadership conversation about AI right now and you hear the same question asked a dozen ways: should we be using this? It sounds like the decision to make. It is really the surface of four smaller ones. “Should we use AI” reads like a technology question, the kind you route to whoever owns the tools, but the moment you try to act on it, it breaks into four much smaller, much more ordinary decisions. None of them is technical. They are decisions you already make every week, about people, work, and judgment.
Here is what the hype obscures. AI fluency for leaders, the skill of working well with these tools, is a leadership skill, not a coding one, and most of it you already have; it has never been laid out as decisions you can recognize. This piece does that. It borrows a ready-made frame, the 4D AI Fluency Framework: Delegation, Description, Discernment, Diligence (developed by Rick Dakan of Ringling College and Joseph Feller of University College Cork, in collaboration with Anthropic, whose free AI Fluency course teaches it in full; used here as a thinking frame, not endorsed as a product), and turns each competency into the real question underneath it.
Each of the four points at a decision you actually face, and each has its own practical companion piece in the series this piece opens: short, concrete pieces with hands-on exercises that teach you to work the decision, not merely nod at it. Read this for the shape of the whole, then follow the tracks to put each one into practice.
Question one: what do I hand over, and what do I keep?
This is Delegation, the first decision because everything else depends on it. Strip the software away and the question is one you answer constantly: which work do I do myself, which do I hand off, and what do I keep my hands on no matter what?
You already do this with people. When you give work to a colleague, you don’t hand over the goal, the judgment, or the accountability. You hand over the doing, and keep the final word. An AI agent is the same decision with one new variable: how much rope to give something that is fast, tireless, occasionally confidently wrong, and not a person you can hold responsible. The honest answer is rarely “all of it” or “none of it.” It is a deliberate split: some work automated, some done together, some reserved for human hands only.
I see this in my own work. With a regulated firm I advise, the person who got the most out of an agent was not the most technical one in the room; it was the operator who knew the rules cold. We put an agent onto the first drafts of routine, high-volume client documents: real volume, real deadlines. What made it work was a subject expert who could say precisely which rule each draft had to satisfy and could catch a plausible-but-wrong answer at a glance. She wrote none of the automation. She owned the standard the output was measured against, and that ownership, not any coding, was the qualification. The agent carried the volume; she kept the judgment and the final sign-off. Command of the work is the skill that transfers; coding is not.
The decision this teaches you to make: where the line sits, this quarter, between what you delegate and what you hold. The Delegation companion, What to hand an agent, and what to keep, works one real, recoverable task end to end, so you finish with a felt sense of that line rather than an opinion about it.
Question two: how do I describe the work so it comes back right?
This is Description, and it is where most disappointment with AI actually comes from. People treat these tools like a search box (type a few words, expect a finished answer), then conclude the technology is overrated when the result is generic. The tools are not search boxes. They are collaborators that do exactly what you describe, which means a vague brief produces vague work, every time.
The fix is not a clever phrase or a secret prompt. It is the discipline you’d use briefing a capable new hire who is brilliant but has no context on your business. You tell them three things: what you want (the outcome and what “good” looks like), how to approach it (the steps, the constraints, the things to avoid), and what kind of partner you need them to be (a fast first draft, or a careful, challenge-everything thinker). Get those three right and the quality of the work changes completely.
That clarity is also the part you can’t outsource. If you cannot describe success to a competent stand-in, no tool will rescue you. That is not a limitation of the tools; it is the leadership skill itself, sharpened.
The decision this teaches you to make: how to turn a fuzzy ask into a brief that produces usable work the first time. The Description track is the deepest in the series, because there is real craft here: from writing one good brief, to setting standing instructions so you never repeat your conventions, to planning a big piece of work before letting anything run.
Question three: how do I know whether what came back is any good?
This is Discernment, and it protects you from the most expensive failure mode: confident, fluent, wrong. AI output reads well even when it is incorrect. The polish is real; the accuracy may not be. A leader who can’t tell the difference, or who hasn’t built a way to check, is gambling, not delegating.
Discernment is judgment applied in three places, and again it mirrors how you already review work. Is the result right (accurate, coherent, fit for who it’s for)? Was the approach sound, or did it loop in circles and quietly smuggle back an idea you’d already rejected? And was the exchange efficient, neither a runaway nor a tool you have to nurse through every step? You don’t need to read code to ask these; you read the work the way you’d read a junior’s draft: skeptically, against what you know to be true.
The practical move is to never accept “it’s done.” You ask the work to prove itself: show the result, flag every claim with no source, cross-check the numbers. Then you approve the outcome, not the activity. Better still, you build the checking in so it happens by default.
The decision this teaches you to make: how to inspect AI output fast enough to trust it and rigorously enough to catch what’s wrong. The Discernment track turns that into a repeatable habit, including the powerful pattern of having a second pass, with fresh eyes, check the first.
Question four: how do I stay responsible while moving fast?
This is Diligence, the one most likely to be skipped. That is exactly why it belongs on the leadership list and not the IT backlog. Diligence is staying in command of the loop: what data goes in, where it lives, who can see it, whether any of this aligns with your obligations to clients, regulators, and your own people. In a European context, with the EU AI Act and data-protection rules shaping the ground, that is not a footnote. It is a standing line item.
There is a second, quieter half that rarely gets named: keeping the skill. If you delegate a kind of thinking entirely and stop doing it yourself, the capability fades — yours and your team’s. Fluency that turns into dependence is a liability, not an asset. The leaders who do this well treat AI like a power tool: enormously useful, used deliberately, with the human capability maintained alongside it. The goal is to build judgment, not to outsource it.
The decision this teaches you to make: which guardrails to set, what to log, what to gate, and how to keep your team’s skills sharp while the tools do more of the work. The Diligence track covers both halves: the governance that keeps you safe, and the practices that keep fluency from hollowing out into reliance.
Start here
Read together, the four are one question asked in sequence: What do I hand over? How do I describe it? How do I judge what comes back? How do I stay responsible? That is AI fluency for leaders: four decisions to get good at, not a technology to evaluate. None require a line of code. All reward what you already do well: defining work clearly, choosing who’s a fit, inspecting results, and owning the outcome.
This piece is the frame. The companion pieces are where you put it to work: one track per question, each built around a real task you can try this week.
Begin with Delegation, because the other three only make sense once you’ve decided what to hand over. Its published companion is live now on the AI Leadership Journal: What to hand an agent, and what to keep. Work through it, then come back for Description, where the real craft begins.
The column is already running. Discernment is live too (Don’t trust the benchmark. Judge the work.), with Description and Diligence to follow. Come back to this page for each; it stays the reference.
The AI Leadership Journal is written by Claudius Gramse. evonomics is the independent AI consultancy helping mid-sized European companies embed AI into the work that actually runs their business — evonomics.eu.
