Teams adopting AI coding tools are merging 98% more pull requests, but review times have jumped 91%, according to a report published by The New Stack. The article, written by engineering platform Aviator, draws on observations from senior engineers and engineering leaders who identify AI code review bottlenecks as a top concern. One study cited in the report found that 77% of engineers now spend less time writing code, redirecting that effort toward verifying AI output instead.
The shift is creating review queues that experienced engineers struggle to clear. Some senior developers now face 15 pull requests daily, each containing roughly 400 lines of code. Across 158 engineers in 28 countries tracked by researcher Annie Vella, the burden is driving resignations: some hope the role will revert to its former shape, while others are leaving the profession entirely. Meanwhile, 31% more pull requests are being merged without any review at all, a pattern the report interprets as engineers voting with their behavior.
The report argues that AI-generated code is harder to review because the reasoning behind it disappears. When a colleague writes code, they can explain the tradeoffs they weighed and the alternatives they discarded; when AI writes it, the reviewer must reverse-engineer intent from a diff alone. The code often passes surface-level checks but harbors problems the report categorizes as "plausible but wrong," "over-engineered," "convention-blind," "confidently hallucinated," or "cargo-cult patterns." According to the authors, three-quarters of review feedback is codifiable, meaning it addresses deterministic rules or execution-testable issues rather than genuine judgment calls.
The report recommends three fixes: codify repetitive feedback into automated checks, preserve the reasoning trail from prompts and agent sessions so reviewers can evaluate intent rather than scan syntax, and measure the verification work that prevents low-quality code from reaching production. It warns that dashboards tracking throughput and cycle time miss the effort senior engineers invest in review. The engineers carrying the review burden aren't complaining, the report states—they're quitting, and leaders focused on lines of code generated won't see the exodus coming until those developers walk out the door. Organizations that fail to adjust the code review process will watch their best engineers leave and find that AI-powered teams ship faster but break more often, the report concludes. The mechanics of knowledge transfer are shifting left alongside code review itself: when reviewers read acceptance criteria instead of diffs, they debate tradeoffs and see how the system is evolving, which is where productive collaboration survives in an AI-augmented workflow.