Lesson 1 of 8

Prompting is engineering

7 min read

The same question gets a brilliant answer one day and a useless one the next, and it feels like luck. It isn't. The good answers almost always came from prompts with more parts — and once you can name those parts, you can build a good one on purpose.

A prompt is a thing with parts

Under the surface, a strong prompt is built from a few named parts — a role (who the AI should be), the task (what to do), context (the facts it needs), a format (the shape you want back), and constraints (the limits). You don't need all of them every time. But each one you add is a decision you make instead of leaving the AI to guess.

A prompt has parts you can name and control: role, task, context, format, constraints. Engineering a prompt means deciding each part on purpose, instead of hoping the AI guesses right.

Every part you skip, it fills in

Leaving a part out doesn't skip it — the AI silently picks the most average version. No format named? You get flowing prose. No audience? It writes for no one in particular. The fix is boring and reliable: read your own prompt back one line at a time, and anywhere you'd have a question, the AI does too. Answer those, and the guessing stops.

An unnamed part isn't skipped — it's guessed. If you'd have a question reading your own prompt, so will the AI; the parts you name are the ones it stops guessing.

The classic checklist is six parts — task, format, topic, tone, context, constraints — but you rarely need all six. Think of them as levers to reach for, not boxes to fill. Naming even two or three usually does most of the work.