If scaling alone stops delivering capability jumps, the field shifts toward algorithmic efficiency, better data, and agentic scaffolding — each with different risk profiles than raw scale. How does the end of simple scaling change the safety landscape? Does it make governance easier or harder?