Uber's uReview AI code reviewer covers over 90% of ~65,000 weekly diffs, with 75% of comments rated useful
Uber built uReview, an in-house AI code reviewer that acts as a second reviewer on its Phabricator diffs across all six monorepos (Go, Java, Android, iOS, TypeScript, Python). A prompt-chained pipeline generates comments with specialised assistants (bugs, Uber best practices, AppSec), then grades, deduplicates and category-filters them to cut false positives. Uber says it analyses over 90% of the ~65,000 weekly diffs landed and reviews each commit within a median of 4 minutes in CI. Engineers mark 75% of its comments useful and over 65% are addressed, against 51% for human-written comments. Uber estimates about 1,500 reviewer hours saved a week. The best setup paired Claude 4 Sonnet as generator with o4-mini-high as grader.
The numbers (as reported)
| Coverage | Over 90% of ~65,000 weekly diffs landed |
|---|---|
| Comments rated useful | 75% (by engineers who interact with the tool) |
| Comments addressed | Over 65% (vs 51% for human-written comments) |
| Review latency | Median 4 minutes per commit in CI |
| Commits processed | Over 10,000 a week, excluding config files |
| Estimated time saved | ~1,500 hours a week (~39 developer years a year) |
| Monorepos | All six |
Facts
- Company / subject
- Uber
- Tools
- Unstated
- Models
- Claude 4 Sonnet, OpenAI o4-mini-high
- In-house tools
- uReview (Uber)
- Codebase
- brownfield
- Work type
- Company rollout
- Stack
- Go, Java, Android, iOS, TypeScript, Python, Phabricator, Apache Kafka, Apache Hive
- Published
- 2025-08-12
- Last checked
- 2026-11-10
What the source claims Practitioner (own blog/paper)
Claimed by the builder, writing about their own work. Not checked independently.
- Analyses over 90% of the ~65,000 diffs landed each week Over 90% of ~65,000 weekly diffs
“uReview today analyzes over 90% of the weekly ~65,000 diffs (equivalent of pull requests) landed at Uber.”
- 75% of comments marked useful and over 65% addressed 75% useful, 65%+ addressed
“Engineers who interact with the tool mark 75% of its comments as useful, and we see over 65% of its posted comments addressed.”
- Addressed at a higher rate than human reviewers' comments (51%) 65% vs 51%
“Internal audits show that only 51% of human-written comments are considered as bugs by the author and addressed in the same changeset.”
- An estimated 1,500 hours saved a week ~1,500 hours a week (estimate)
“This translates to approximately 1,500 hours saved weekly, equivalent to nearly 39 developer years annually.”
Caveats
Self-reported by Uber. This is an AI code reviewer, not an agent that writes code, and the post gives no merge-rate, cycle-time or defect outcomes. It does not say which coding agents authored the reviewed diffs, so the tool is Unstated. Published August 2025, earlier than Uber's other stories on the site. Volume figures are inconsistent: '~65,000' diffs is called weekly in one place and 'per month' in another, and 'over 90%' coverage sits awkwardly with 'over 10,000 commits' processed a week. The 1,500 hours a week is an estimate assuming a second human reviewer needs 10 minutes per commit. The usefulness rate counts only engineers who rate comments. Fixer, the companion tool that proposes code changes, is not covered.
More stories by tool: Unstated
Related topics: Company rollouts and costBrownfield codebases
Source
Uber Engineering blog (Sonal Mahajan, Shauvik Roy Choudhary, Will Bond)