Le goulot d'étranglement, c'est moi
Summary
The article is a reflective piece on the bottlenecks of running multiple AI code agents across projects. It discusses attention fatigue, notification overload, context window compaction, and persistent locks, and proposes practical patterns like externalizing the prompt, selective alerts, and a control tower concept for agent orchestration. The piece also hints at building an orchestration tool for agent-based workflows.