Context engineering is the discipline of deciding exactly what information to place in a language model’s context window to get the best possible answer. Because a model can only see what fits in its limited context, and because its attention degrades when that context is cluttered, choosing what to include is a high-leverage design problem.
It goes well beyond simply retrieving the top few chunks. Context engineering involves selecting the most relevant evidence, summarising or compressing long material, ordering information so the most important parts are prominent, trimming redundancy, and formatting everything so the model can use it. The goal is maximum useful signal per token.
This matters because more context is not automatically better. Models exhibit a lost in the middle effect, paying less attention to information buried in the centre of a long prompt, and every extra token adds cost and latency. Precise, well-curated context routinely outperforms simply stuffing in everything that might be relevant — which is why context engineering has become a core skill in building reliable AI applications.