Prompt engineering

Writing your instructions to an AI deliberately rather than off the cuff: telling it who to be, what the rules are, showing it one example of a good answer, and saying what shape you want the result in. The raw material is the prompt.

There's nothing mystical about it and no special language to learn. It's the difference between telling a new hire "sort this out" and handing them a one-page brief with last month's version attached.

Why it matters: the gap between a mediocre AI result and an excellent one is usually the brief, not the AI. It's also the most learnable thing in this entire dictionary — no technical background needed, and you already do it every time you delegate well.

Worth less than it used to be, though. Models have got much better at working out what you meant from a plainly worded request, so clever phrasing matters less every year. What you put around the request — the right documents, in the right order, with the wrong ones left out — now matters more. That's context engineering, and it's where the effort has moved.

Related: Prompt · Context engineering