Nate B Jones gives both agents access to the same problem-discovery skill and asks them to inspect a working repository. The results reveal different priorities: Fable focuses on product and workflow pain, while Codex gravitates toward defects it can concretely verify and fix.
Nate B Jones argues that the comparison exposes a hidden source of agent quality. Two capable systems can produce useful work while optimizing for different definitions of importance, so asking several agents to diagnose the same environment can uncover a broader opportunity set.
Nate B Jones recommends matching the model to the stage of work. Exploratory systems can frame and rank problems, then clearly scoped implementation tasks can move to agents that are dependable at finishing them.
Watch the original on YouTube