Analytics covers what your organisation is producing and what it costs. You’ll find it under Analytics, with an Executions view for individual runs.
Throughput
How much work is completed over time.
Assess it against queue depth, not in isolation. Flat throughput with an empty queue means demand is steady. Flat throughput with a growing queue indicates a capacity limit — a different scenario requiring a different fix.
Execution history
Every run, including its duration, iterations and token usage.
Iterations provide the most useful diagnostic figure. A task needing many iterations faced difficulties — perhaps due to vague specifications, a complex codebase, or repeated rejections during verification. Duration alone cannot differentiate these cases from situations where simply a large amount of work was involved.
If a task costs more than expected, check its execution history before drawing any conclusions about model pricing.
The three key ratios to monitor
Rework rate. What proportion of tasks return to the developer after review. A rising rework rate usually stems from poor specification quality — see Specifications.
Queue time versus execution time. Which one predominates indicates whether you need to increase capacity or adjust how you scope work.
Spend versus merged work. This is the true measure of efficiency. Rising spend alongside delivery shows the system is functioning properly; rising spend without corresponding delivery means resources are being wasted.
Attributing cost
Regarding spend specifically, the billing views break down usage by project and agent — see Usage and reports. One project often accounts for a disproportionately large share, making it the best area to focus on for improvements.
Receiving this information automatically
To get a regular summary sent to your email or chat channel instead of having to remember to open it yourself, see Reports.