GLOSSARY
Context Engineering
The discipline beyond prompt-writing: deciding what goes into a model's limited context window — instructions, retrieved documents, tool results, history — and in what order.
A prompt is one ingredient; the context window is the whole kitchen. Context engineering is the systems work of deciding what earns a slot: compressing conversation history, selecting which retrieved passages to include, structuring tool outputs, and ordering material so instructions survive long conversations. As agents run for hours, managing this budget becomes the difference between coherent and derailing behavior.
The name marks a maturity shift: teams realized that models fail less from bad wording than from bad inputs — too much noise, missing facts, stale history. Practical levers include retrieval quality (reranking), summarization checkpoints, context-window triage, and the newer trick of letting models request context on demand rather than front-loading everything.