Context engineering

Context engineering is the practice of deciding what information an AI model sees before it acts: the right files, the relevant rules, the prior decisions, the examples, and nothing else. A model only knows what is in its context window, so the quality of its output is capped by the quality of what you put in front of it.

This has overtaken prompt engineering as the skill that separates good AI builds from frustrating ones. Give a coding agent your whole repo and it drowns; give it the three files that matter plus a clear instruction and it nails the task. You are not writing more code, you are curating what the model reasons from: stable conventions, a memory of locked decisions, the right reference at the right moment, so every task starts from the same solid ground instead of the model guessing from scratch.

In practice it looks like this. Say you're building a feature with Claude as your coding agent: instead of pointing it at the whole codebase, you hand it the two files it will edit plus your conventions doc, and it stops inventing patterns you don't use. Say you're researching a dense topic and load the ten source PDFs into Notebook LM so its answers are grounded only in those documents, not the open web. Or say your team's decisions and specs live in Linear: pasting the relevant issue and its locked acceptance criteria into the model's context is what stops it re-litigating a call you already made. Same model, far better output, because you curated what it saw.