glossary
Context Engineering
The discipline of controlling which context, in what volume and at what priority is supplied to an AI-workflow step.
Context engineering is the discipline responsible for which context, in what volume and at what priority reaches a specific step of an AI workflow. It covers re-indexing on source-change events, re-ranking of what was retrieved, per-step context budgets, and evaluating which fragment actually influenced the answer.
How it differs from the naive version: “retrieve similar and put it all in the prompt” seems sufficient, but on long noisy context the model extracts the essential worse and costs more. Context engineering is the decision “what of the retrieved is actually needed”, made by the system rather than offloaded to the model.
Why it matters: answer quality depends more on context relevance than on model size, and even than on context completeness. It is a distinct engineering discipline — without it, knowledge-retrieval systems degrade invisibly.