I kept hitting the same workflow failure mode in complex Codex tasks:
delegate the work → the main agent gets impatient → checks too early → decides the workflow is stuck → quietly takes the task back.
The issue wasn’t lack of agents. It was blurry ownership.
So I built
