The U-turn is more telling than the restriction itself: a CIO openly celebrating vibe coding in July and walking it back within months is a textbook case of AI evangelism colliding with regulated reality. The stated rationale — productivity and risk management — reads like compliance catching up with enthusiasm, and in banking that check always arrives eventually. In our view this is less a rejection of AI coding tools than a maturing of their deployment: engineers with review and audit discipline keep access, while ad-hoc scripts written by traders and bankers become liability. Expect other banks and large enterprises to converge on the same split, because 'anyone can write code' and 'anyone should ship code' are very different claims. The real signal is that governance, not model capability, is now the bottleneck for enterprise AI adoption.

Rao's mercantilism analogy is sharper than the usual job-loss panic: scarcity-thinking treats minds as stockpiles rather than engines of demand. We find the historical parallel instructive but incomplete — Smith's rebuttal to mercantilism took decades and real human misery, so even if demand for human thinking grows, the transition won't be gentle. The deeper flaw: AI competes directly with cognition, unlike steam replacing muscle, so whether humans keep comparative advantage in judgment, taste and accountability remains an open question. Where the concept lands hardest is policy — chip export controls, closed-weight regimes and licensing debates already look quite mercantilist. Read it as a lens, not a forecast.