Analytics covers what your organization is producing and what it costs. Find it under Analytics, with an Executions view for individual runs.
Throughput
How much work is completing over time.
Read it against queue depth, not on its own. Flat throughput with an empty queue means demand is flat. Flat throughput with a growing queue means you are capacity limited — a different situation with a different fix.
Execution history
Every run, with its duration, iterations and token usage.
Iterations are the most diagnostic number available. A task that took many iterations struggled — it was under-specified, the codebase was hard to navigate, or verification kept sending it back. Duration alone cannot distinguish those from "there was simply a lot to do".
When a task costs more than expected, open its execution history before concluding anything about model pricing.
The three ratios worth tracking
Rework rate. What proportion of tasks go back to the developer after review. Rising rework almost always traces to specification quality — see Specifications.
Queue time against execution time. Which one dominates tells you whether to buy capacity or change how you scope work.
Spend against merged work. The honest efficiency measure. Spend rising alongside delivery is the system working; spend rising without it means something is being wasted.
Attributing cost
For spend specifically, the billing views break usage down by project and agent — see Usage and reports. One project is usually responsible for a disproportionate share, and that is where a fix pays back most.
Getting this pushed to you
For a recurring summary delivered to email or a chat channel rather than something you remember to open, see Reports.