He’s right. The interesting question is who lands on which side of that line.

Two engineers feed Claude the same prompt. The output looks plausible. The tests pass. One ships it. The other notices in thirty seconds that the model quietly removed a retry-on-rate-limit pattern the test suite didn’t cover, says no, regenerates.

Same tool. Same prompt. One has the understanding. One doesn’t.

The split: engineers who built the understanding before AI got cheap, and engineers who treated cheap AI as a substitute for building it.

Without that muscle, AI doesn’t multiply you. It just ships your blind spots faster.

That’s the actual filter.

Longer version here: https://joaofogoncalves.com/articles/2026/05/2026-05-04-ai-as-the-great-filter/