The practice of designing dynamic systems that provide optimal information, tools, and format to LLM context windows, evolving beyond simple prompt engineering to enable reliable AI agent performance through techniques like compaction, structured note-taking, and multi-agent architectures.
