What Would Good Agent Productivity Metrics Look Like?

I'm curious how others are thinking about productivity as you're adopting AI agents. We're starting to do some deep analysis of the output/logs from agent tooling to look for the new bottlenecks.