You are not behind on AI. You are stuck in the prompt age.

We see it constantly, and not where you'd expect. The people stuck are often the most capable in the room — senior engineers, partners at consultancies, operators who run real businesses. They are fluent in their domain and completely confident with a chatbox. And they have quietly plateaued.

The pattern is always the same. They have learned to prompt. They can coax a decent answer out of a model. So they assume that is the skill, and that they have it. We call this AI illiteracy, and it is the most expensive kind, because it is invisible from the inside.

Prompting is roughly a quarter of one of four competencies. Even world-class prompting addresses about 25% of what working with AI actually requires.

That framing comes from Anthropic's AI Fluency work, developed with academics Joseph Feller and Rick Dakan. They define four competencies — the 4Ds — that together describe what it means to work with AI effectively, efficiently, ethically and safely. Prompting sits inside just one of them.

The four things, briefly

• Delegation — deciding what to hand to AI at all, and what stays human.
• Description — communicating the work clearly. (Prompting lives here.)
• Discernment — judging whether the output is actually any good.
• Diligence — owning the result and how it was produced.

The prompt-age person has spent all their energy on the second one and none on the other three. They delegate badly — handing over things AI shouldn't touch, or refusing to hand over things it would do brilliantly. They rubber-stamp output instead of discerning it. And they rarely think about accountability until something goes wrong.

What breaking through looks like

Anthropic's own research found engineers describing their job shifting to "manager of AI agents" — one estimated 70%+ of their work had become reviewing and revising rather than writing from scratch. That is not better prompting. That is a different mode of working entirely: delegate the work, describe it well, discern the result hard, stay diligent about what ships.

The good news, as Feller puts it, is that this is empowering rather than threatening. Fluency restores your agency over how AI gets used. It is also learnable — and it is independent of whichever model happens to be current. You are not behind. You have just been practising one quarter of the skill.