Type: url
Original: https://rlancemartin.github.io/2025/06/23/context_engineering/
Status: generated
Summary
A blog post by Lance Martin (June 23, 2025) that frames context engineering as filling an agent’s limited context window with just the right information at each step, and groups the strategies seen across popular agents into four buckets: writing context outside the window (scratchpads and memories), selecting it into the window (memories, tools and knowledge), compressing it (summarization and trimming), and isolating it (sub-agents, sandboxed environments and runtime state). Each bucket is illustrated with examples from products, papers and agent frameworks.