Context engineering
Choosing everything an AI sees for a task (instructions, files, past messages, tool results), not just the wording of the request.
A prompt is one message. But by the time a model answers, it's reading much more: the app's system prompt, your custom instructions, memories, uploaded files, search results and the whole conversation so far. Context engineering is deciding what goes into that pile.
The word took off with AI agents, which run for many steps and keep collecting text as they go. Anthropic describes the goal as finding "the smallest set of high-signal tokens" that gets the result you want. In plain terms: everything the task needs, and as little else as possible.
For everyday use it comes down to habits. Give the relevant extract, not the whole archive. Start fresh when a chat gets long. Keep separate projects for separate jobs.
For example
Instead of pasting a whole 200-page manual, you give the AI the 3 relevant chapters and a one-line note on what you're deciding.
Put it to use
Where you'll meet context engineering in ChatGPT, Gemini and Claude, how to use it there, and the mix-up most people make.
Daily Bytes Plus is in the Android app.