DORA still rules, reviews are the choke point
DORA metrics remain the core way to measure delivery performance, but engineers report that AI code generation increases volume without shortening lead times because code review becomes the bottleneck. Observers point to review speed and quality as the missing investment if teams want AI to translate into faster, safer deployments (x.com/twtayaan/status/2042259993005543925; x.com/vmblog/status/2041842494283841698).
A team can now generate a week’s worth of code in a day, then watch it sit in a pull request queue for five days waiting for a human to read it. That is why software teams keep talking about speed while their lead time barely moves. (dora.dev) (faros.ai) The yardstick most teams still use is DevOps Research and Assessment, a research program usually shortened to DORA. Its core measures ask four plain questions: how often you deploy, how long a change takes to reach production, how often releases break, and how fast you recover. (dora.dev) (docs.gitlab.com) Those measures matter because they track the whole trip, not just typing speed. A code assistant can make the first mile faster, but DORA lead time still includes review, testing, approval, and deployment before users see anything. (dora.dev) (docs.gitlab.com) That is where the new pileup shows up. Faros AI said in July 2025 that teams with high artificial intelligence adoption completed 21% more tasks and merged 98% more pull requests, but pull request review time rose 91%. (faros.ai) A pull request is the packet of changes one engineer asks another engineer to inspect before merging. If artificial intelligence doubles the number of packets and senior reviewers do not double with it, the queue gets longer even if every coder feels faster. (faros.ai) SonarSource described the same shift with different numbers. Its 2025 survey, based on more than 1,100 developers worldwide, said 96% of developers do not fully trust the functional accuracy of artificial intelligence generated code, so the work moves from writing to verifying and debugging. (events.sonarsource.com) Harness found the bottleneck extends past review into the rest of the delivery line. In its 2025 survey of 900 engineers and managers across four countries, 63% said they were shipping faster with artificial intelligence, but automation averaged 51% in coding workflows and only 43% in continuous integration and build pipeline work. (harness.io) That mismatch explains why DORA still rules the conversation. Deployment frequency can rise for a few teams, but if review queues swell and broken releases increase, change failure rate and time to restore service can erase the gain. (dora.dev) (docs.gitlab.com) Tool vendors are already moving toward the choke point instead of the keyboard. GitHub now offers Copilot code review on pull requests, including automatic reviews in repositories where organizations turn the policy on, because the scarce resource is no longer code generation alone. (docs.github.com) The practical shift inside engineering teams is simple and expensive: smaller pull requests, stricter review rules, more automated tests, and more reviewer capacity. If the review lane stays one car wide, adding faster engines upstream just creates a longer traffic jam. (faros.ai) (harness.io)
