Context engineering

Deciding what the AI gets to see before it answers: which documents go in, in what order, and above all what stays out. The prompt is the question you ask. Context engineering is everything you put on the desk beside it.

Side by side on one job. Writing a quote for a customer: prompt engineering is wording the request well — "write a quote in our usual style, itemised, VAT separate." Context engineering is choosing to hand over last month's quote for the same kind of job and the current price list, and not handing over the other two hundred quotes in the folder. Identical question either way. Completely different answer.

Why it matters: models answer from what they're shown, and only from what they're shown. Ten relevant paragraphs beat a hundred pages — more isn't better, relevant is better. There's also a hard ceiling: everything you hand over has to fit inside the context window (how much the AI can hold in mind at once), so deciding what to leave out isn't tidiness, it's the job.

Related: Prompt engineering · Context window