Context engineering determines what an artificial intelligence system can see and use when producing an answer or taking action. It includes system instructions, task details, retrieved documents, examples, tool descriptions, memory, permissions, and current application state.
Effective context is relevant, trustworthy, bounded, and structured for the task. More context is not always better. Designers must manage source authority, freshness, contradictions, prompt injection, context-window limits, and the order in which evidence is presented.
