AI coding · Advanced

Context Engineering

Also called: Context design, Managing context, What you feed the model

Deliberately choose and organize what goes into the model context—code, rules, retrieval hits, history—so the AI does the right work inside a limited window.

In detail

Context engineering is less about one clever prompt sentence and more about what sits on the whole "desk": system prompt, rules files, relevant source, errors, design notes, retrieved chunks. The window is finite; the wrong pile makes the model distracted or forgetful.

In practice: @-mention exact files instead of pasting the repo; put stable agreements in rules files; start a new thread after a phase and attach a summary; use RAG for only the relevant slices. It ties to Prompt, Context Window, and RAG, but the lens is system design, not a single question.

A common myth is "more context is always better." Irrelevant code and contradictory instructions lower success rates.

Developer infoTerm ID, DOM cues, match priority

Term ID: ai-context-engineering
DOM selectors: No DOM cues. This concept isn't detected directly on a page.
Priority: 1 · when several match at the same level, the higher priority wins
Version: v1 · updated Oct 5, 2026