@mattsheffield Just because human code is generally not very good, doesn’t mean that vibecoding is “about the same”, because their failure modes are very different.
The way human code fails can be predicted by individual or team habits, quirks, or blind spots; in other words, the more you know a person, the more you know how their code tends to fail. This causes the failure mode to cluster around some weak points, which the human can correct for and improve over time.
Vibe code fails in random, bizarre ways, often exhibiting errors in plausible-looking but ill-advised patterns, or code that looks right in isolation but makes no sense when combined with others. This kind of failure mode is diffuse and stochastic, and is very difficult to correct for.
Good employers don’t necessarily look for *language* competence; they look for *engineering* competence, which involves knowing what you don’t know, and compensating for it using checklists, tests, and (gasp) asking help from other competent people.