Coding agents do better work in some repositories than others, and the difference is mostly things that help human contributors too: one command that runs the agreed checks, fast feedback, clear ownership boundaries and short, specific instructions.
Read ArticleTreat the context window like memory in a constrained system: allocate it, measure it and reclaim it. A per-component budget, a compaction policy that runs in code, just-in-time retrieval, and when to hand work to a sub-agent with a clean context.
Read ArticleConnecting agents through events decouples them, and also exposes two failure modes that ordinary services also face: duplicate delivery of expensive, side-effecting work, and agents triggering each other in loops. An event envelope and handler guards for both.
Read Article