Estimate how many AI code reviews a serving pool can complete over time and whether it can keep up with incoming work. The calculator starts with aggregate token throughput, adjusts it for target utilization and service availability, and divides effective capacity by the average token volume of one review. It then compares capacity with demand and estimates queue growth or spare capacity. Use workload-weighted token measurements for mixed repositories.
Effective tokens/sec = workers × tokens/sec/worker × utilization × availability
Reviews/minute = effective tokens/sec × 60 ÷ tokens/review
Daily capacity = reviews/minute × 60 × operating hours
The main result is sustainable completed-review capacity under the entered utilization and availability assumptions. A negative demand margin indicates an accumulating queue if the arrival rate remains constant.
Availability is applied as expected productive time. This does not reproduce bursty arrivals, request-size variance, batching effects, or percentile queue delay.
Eight workers at 150 tokens/second, 75% utilization, and 99% availability provide 891 effective tokens/second. With 10,000 tokens per review, capacity is 891 × 60 ÷ 10,000 = 5.35 reviews per minute.
The target leaves capacity for variation, coordination, and bursts instead of assuming workers stay at maximum load.
It proportionally reduces effective throughput to represent expected unavailable time.
Incoming work exceeds sustainable completion capacity, so unfinished reviews accumulate while that condition persists.
Yes. Per-minute capacity still calculates, but daily completed capacity becomes zero.
Only if human time occupies the same constrained serving workers; otherwise model it as a separate workflow stage.
| Measure | Unit | Time basis |
|---|---|---|
| Worker throughput | tokens/second | Active serving |
| Review size | tokens/review | Average workload |
| Capacity | reviews/minute | Sustainable estimate |
| Daily capacity | reviews/day | Entered operating hours |