What is Context engineering?

Context engineering is the discipline that grew out of prompt engineering once agents started pulling in data from many sources. Where prompt engineering focused on wording a single instruction well, context engineering is the broader job of assembling the entire context window: which system instructions to set, which tools to expose, which documents and prior results to retrieve, what to summarize, and what to leave out. The guiding insight is that more context is not better, the right context is. Irrelevant material wastes the token budget, raises cost and latency, and can actively degrade answers by burying the signal. So practitioners think about selection (retrieve only what is relevant), compression (summarize long histories), ordering (place key facts where the model attends best), and isolation (give sub-agents their own focused contexts). Memory systems are a core tool here: instead of replaying an entire conversation history, a well-designed memory layer surfaces just the few observations that bear on the task. Glen is built around this idea, its read pipeline scores stored observations for relevance and returns a tight, task-specific context block rather than a data dump.