Stop Counting PRs. Measure the Queue.

Описание и применение

Перейдите от подсчёта PR к измерению очереди: время ожидания, входящие PR, размер пакета, первая одобрённость, повторная работа, покрытие контрольных точек и стоимость одобренных изменений.

Задача

Teams using AI agents generate code faster than they can review it, leading to inflated PR counts that do not reflect true productivity.

Результат

By shifting focus from PR counts to queue metrics, teams can identify review bottlenecks, ensure quality of agent output, and align productivity with actual delivery outcomes.

Рабочий процесс

• Track time from PR creation to first human review.
• Record incoming PRs and PR size each week.
• Monitor batch size to detect increasing review complexity.
• Measure first-pass acceptance and rework rate of agent-authored code.
• Track harness coverage for known failure modes.
• Calculate cost per accepted change including compute, tokens, and review time.
• Compare queue metrics against DORA metrics (deployment frequency, lead time, change failure rate, time to restore).
• Adjust review capacity or agent output based on queue growth patterns.

Доказательства

AI agents can generate code faster than your team can review it.

Источник

Instead of counting pull requests alone, track: • Review wait time: Time from PR creation to first human review.

Источник

Track a few additional metrics: • First-pass acceptance: How often agent changes are accepted without human edits.

Источник

Keep the four DORA metrics, too: deployment frequency, lead time for changes, change failure rate, and time to restore service.

Источник