Context Engineering
Context engineering is the discipline of deciding what information goes into a model's context window for a task — which documents, which history, which tool output — and how fresh and trustworthy it is. As models commoditize, it's where a lot of the durable advantage now lives.
Also known as: context engineering
Prompt engineering is how you word the request; context engineering is everything you put in front of the model around it — the retrieved documents, the conversation so far, the tool results, and the rules for what’s allowed in. It treats the system around the model as real software to design, not a magic box to prompt.
It matters more every quarter because the models themselves are converging. When everyone has access to similar frontier models, the differentiator is what you feed them and how well you manage it — freshness, relevance, governance, and what you deliberately leave out. Done badly, you get context poisoning and bloated cost; done well, it’s the part of the stack you actually control.
