Context Engineering
Systematically constructing and evolving an agent’s context — instructions, memory, retrieved info, tool results — as the primary lever for performance.
Context engineering reframes prompt design as managing the full evolving context window — playbooks, memory, and curated retrievals — that an agent carries across steps.
It became a named discipline in 2025 as agents grew long-horizon; methods like Agentic Context Engineering formalized it with incremental delta updates and reflect/curate loops to let models self-improve at test time without weight updates.