[!todo] Seed note. A starting point, not a finished note yet.
Context engineering is the successor to prompt engineering for agentic coding: instead of crafting one clever message, you build durable infrastructure (spec files, context files like AGENTS.md, plan and progress artifacts) that shapes agent behavior across sessions and projects. It matters because models have finite context windows and attention budgets, so what you feed and when you feed it determines output quality more than raw model capability at the margin. The failure modes are symmetric: too little context and the model invents defaults, too much and its attention degrades. Seeded from Upfront Planning in the AI Era - PRDs, Design Docs, and Specs; closely related to AGENTS.md.