AI fluency is the ability to use AI well in your own work: knowing which jobs to hand to it, how to ask for what you want, what it needs to see, and how to check what it gives back before you rely on it. It is a set of habits, not a set of facts. You measure it by watching what people do with AI in real situations, not by counting the hours of training they have watched.
What fluency looks like at work
Think of a colleague who is good with AI. What do they actually do differently? Usually five things:
- They pick the right jobs. They hand AI the drafts, summaries and rewording it does well, and keep the jobs where one wrong detail would do harm.
- They brief it clearly. They say what they want back, who it is for, what the limits are and what counts as done. So they rarely need three rounds of edits.
- They give it the right context. They show it the document or figures it needs, leave out what it shouldn’t see, and know that a new chat may not remember the last one.
- They check before they send. They read the answer against what they know and catch the confident mistake. AI can state something false as if it were true; the industry calls this a “hallucination.”
- They keep what works. When a job repeats, they save the instructions so they don’t start from scratch each time.
None of this needs code. It needs judgment, and judgment is built by practice.
One published framework
Anthropic, the company that makes Claude, publishes a free course on this, AI Fluency: Framework & Foundations, written with Prof. Joseph Feller (University College Cork) and Prof. Rick Dakan (Ringling College). It organizes fluency into four parts it calls Delegation, Description, Discernment and Diligence. Roughly: deciding what to hand over, describing it well, judging what comes back, and taking responsibility for how it is used. It lines up closely with the habits above, and the course is worth taking.
Why hours and completions don’t measure it
Most training is measured by what is easy to count: hours watched, modules finished, quiz scores. These tell you someone was present. They don’t tell you whether that person now briefs AI well or catches its mistakes.
Take a quiz question like “What is a hallucination?” Someone can answer it perfectly and still forward an AI summary with an invented figure in it. Knowing the word and catching the error are different skills. Only the second one protects your work.
Self-ratings have the same problem. People who have used AI a little often rate themselves highly, because they haven’t yet met the cases where it fails.
How to measure it: watch decisions
To see fluency, put a person in front of a realistic situation and ask for a decision. Then judge the decision, not the wording.
- Judgment: Here are six tasks from a Monday to-do list. Which go to AI, which need checking first, and which need a different tool?
- Briefing: Here is a vague request. Rewrite it so that someone who follows instructions to the letter gets it right the first time.
- Context: Here are five documents. Which does the AI need to see for this job, and which should stay out?
- Checking: Here is an AI draft and the facts it was based on. What doesn’t hold up?
A few rules make this kind of measure honest:
- Count the first try at a new situation. If someone has seen the exact case before, a right answer may just be memory. Fluency shows when the situation is new.
- Change the situation. Someone who gets it right in two different settings is more likely to have the habit than someone who got one case right.
- Grade by fixed rules. The same answer should get the same result for everyone, every time.
- Build in order. Checking an AI’s reasoning depends on knowing what it can see. Don’t credit an advanced skill before the basic one it rests on.
Levels, not pass or fail
Fluency grows in steps, so a single pass or fail hides most of what you want to know. A simple ladder of levels works better. One way to name them:
- Aware: can see it. Recognizes the situation, and tells a good example from a bad one.
- Practicing: can do it when prompted, in a situation that is laid out with some guidance.
- Capable: can adjust it when the inputs, the limits or the setting change.
- Independent: notices when it is needed and does it without being told how.
- Leading: takes it further: builds it into how the team works, teaches it, improves it.
Give each person a level on each habit, rather than one overall score. Someone can brief AI very well and still be weak at checking its work. That is exactly what a manager needs to know.
What to do with the result
A measure is only useful if it changes what happens next. For yourself, it shows which habit to practice. For a team, it shows where people are stuck. If several people miss the same kind of decision, that is a topic for the next team meeting, not a reason to rank anyone. For more, see how a company should train employees on AI.