How do you review AI-written code without slowing teams down?

Summary of public sourcesWritten by our research agent from 6 public sources. Not first-hand experience.

AI raises PR volume while review time does not shrink. In the 2025 Stack Overflow survey, 66% of developers named answers that are almost right as their top frustration. 45.2% said debugging AI-generated code takes more time. Practitioners describe verbose code with no clear design thread, and reviews that have grown thinner.

The most common answer is that the author owns the PR. Bitly requires the human author to review AI code before it goes to review. Teams lean on linters, tests and secret scanning so humans can focus on intent and design. Many pre-screen with an AI reviewer. Uber built one that runs on over 90% of its roughly 65,000 weekly diffs. It filters out low-confidence and low-value comments, and engineers rate 75% of its comments as useful. Uber still leaves design and architecture to humans. Practitioners also scale scrutiny by risk and want reviewers to see the agent's plan, not only the diff.

Still open: whether AI reviewing AI code can be trusted on changes that matter.

Sources

• Hacker News: Ask HN: How is your org managing PR review load as AI multiplies code output?
• Hacker News: Ask HN: What tools are you using for human code review of AI-assisted code?
• Survey: AI | 2025 Stack Overflow Developer Survey
• Engineering blog: uReview: Scalable, Trustworthy GenAI for Code Review at Uber
• Engineering blog: How We Rolled Out AI Coding Tools at Bitly: Six Months and Counting
• Reddit: AI generated code review