Addy Osmani on managing coding agents with self-verifying tasks

Addy Osmani says coding agents turn developers into managers, so the goal is to let agents verify their own work rather than running more parallel sessions.

He recommends giving each task executable checks such as tests, builds, type checks, and screenshots for UI changes. He also says human attention, not agent capacity, is the main limit on parallel work, so unattended tasks need clear completion criteria, permissions, and review. The source is a post by meng shao on X summarizing Osmani's advice.