Span can be useful when you need to understand AI’s impact and costs within engineering. It analyzes agent traces, ties them to pull requests (PRs) and delivery outcomes, attributes AI-assisted code, and tracks AI usage and costs by team, tool, and model.