Running AI well is less about prompts and more about process. The people getting real work out of these tools treat them like a capable teammate who has no memory: they put the important state somewhere durable, they break work into steps a fresh session can pick up, and they keep a human in the loop on the calls that matter.
Three habits do most of the work. First, write things down where the tool can read them — a brief, a plan, a doc — so you are not re-explaining context every session. Second, separate the steps that need judgement from the steps that are just typing, and let the tool own the typing. Third, verify: never ship what you cannot check.
This page is a living summary. It will change as the practice does — the history below shows how the thinking has moved.
How this has evolved
June 14, 2026
First cut — the teammate framing.
Read this versionHide
Getting useful work out of AI starts with treating it like a teammate, not a search box. Give it context, ask for one thing at a time, and check what comes back before you rely on it.