Where prompt engineering optimizes what you ask, context engineering optimizes what the model can see when it answers: which files to include, when to summarize or compact history, what to retrieve versus what to keep resident. It's especially central to long-running coding agents, where context window budget is a scarce, actively managed resource.