Delivery · Report
AI doubled the pull requests. Now review is where the work waits
AI made writing code cheaper. It did not make reading it cheaper. Two years of delivery telemetry show where the extra work goes: into the review queue.
By Nythrex team2 min read
Every engineering lead knows the shape of the week now. Monday’s tickets turn into Tuesday’s pull requests faster than ever, and by Thursday the same three senior engineers are buried in review requests. The code was written in an afternoon. It waits for days.
This is not just a feeling. Faros AI, which collects delivery telemetry from engineering organisations, has published two reports on what happens as AI adoption rises. Read together, they describe a bottleneck that has moved rather than disappeared.
- pull requests merged on teams with high AI adoption — Faros AI, 2025
- +98%
- pull request review time on the same teams — Faros AI, 2025
- +91%
- average pull request size — Faros AI, 2025
- +154%
- bugs per developer at high AI adoption — Faros AI, 2026
- +54%
What the telemetry says
The first Faros report, published in July 2025, covered more than 10,000 developers on 1,255 teams. Developers on teams with high AI adoption completed 21% more tasks and merged 98% more pull requests. Review time on those teams rose 91%, and pull requests grew 154% larger. At company level, Faros found no significant correlation between AI adoption and overall throughput, DORA metrics or quality.
The 2026 follow-up, The Acceleration Whiplash, widened the sample to about 22,000 developers on more than 4,000 teams. This time throughput did show up — epics completed per developer rose 66%. So did the cost: median time to first review up 156.6%, pull requests merged with no review at all up 31.3%, bugs per developer up 54% and incidents per merged pull request up 242.7%.
AI made writing code cheaper. It did not make reading it cheaper.
Why review became the queue
Review time scales with the size of a change, not with how it was produced. A pull request two and a half times as large takes longer to read, and an AI-written one still needs a person who understands the system to approve it. Those people are few, senior and already busy.
When the queue grows, teams cut corners in predictable ways: rubber-stamp approvals, merging without review, bigger batches to save reviewer time. The 2026 numbers — more unreviewed merges, more bugs, more incidents per change — are what those shortcuts look like in aggregate.
What to do about it
Treat review capacity as the constraint it is. Keep AI-written pull requests small enough to review in one sitting. Route them to reviewers who have the context. Track time to first review next to merge counts, and make “merged without review” a number someone looks at every week. None of this needs a tool; it needs the queue to be visible.
Sources
Frequently asked questions
Does AI slow down code review?
It increases the amount to review. Faros AI found review time rose 91% and pull requests grew 154% larger on teams with high AI adoption in 2025; its 2026 report found median time to first review up 156.6%.
Why don’t AI productivity gains show up at company level?
Gains at the keyboard are absorbed downstream, in review, testing and release. Faros AI’s 2025 report found no significant link between AI adoption and company-level throughput; its 2026 report found throughput gains alongside more bugs and incidents.
How do you fix a code review bottleneck?
Keep pull requests small, route them to reviewers with context, track time to first review and unreviewed merges every week, and treat reviewer time as the capacity limit it is.
Read next
- Assign the issue, get a pull request. What does an agent’s PR really cost?Coding agents now take an issue and return a pull request. Two new studies of tens of thousands of agent PRs show where the cost hides: review time, follow-up fixes and reverts.Read
- DORA puts a price on AI-assisted development: the J-curve, the verification tax and the real drivers of returnGoogle Cloud’s DORA team published a framework for estimating the financial return of AI in software delivery. Its message: AI amplifies the system it lands in, and the first months may cost more than they return.Read
- How Nythrex works: from AI session to productionHow Nythrex links the task, AI session, commit, pull request, CI run, deployment and cloud into one unit of work with its own cost and the basis for every figure.Read
Stop guessing.
Start seeing.
No card and no subscription: paid plans have not launched yet.