Context engineering
The craft of choosing what enters a model's context window, files, rules, docs, history, so it reasons about your real system.
A language model only knows what is in its context window. Context engineering is the craft of choosing what goes there. Which files. Which conventions. Which past decisions. Which test output. Done well, the model acts like an engineer who has read the codebase. Done badly, it writes fluent code for a project that does not exist.
The term replaced prompt engineering as the working word around 2024 and 2025. By then it was clear the bottleneck was picking the right facts rather than the wording. Files like AGENTS.md or CLAUDE.md make the craft durable. The rules live in the repo, so every agent session starts from the same ground truth.
Context is also a budget. Windows are finite. Models weight recent tokens unevenly. So the craft is as much about what to leave out, stale docs, dead rules, noise, as what to put in.