Quality assurance · Production

uReview: Scalable, Trustworthy GenAI for Code Review at Uber

The problem

Uber's code reviewers were overloaded by the rising volume of changes driven by AI-assisted development, leaving insufficient time to catch subtle bugs, security vulnerabilities, and best-practice violations—leading to missed errors, slower feedback loops, production incidents, and wasted resources.

First attempt

Third-party AI code-review tools were evaluated and found unsuitable: most required GitHub (Uber uses Phabricator), suffered from many false positives and low-value true positives, and could not interact with Uber's internal systems. Their per-diff costs at Uber's scale were also an order of magnitude higher than running uReview in-house.

Workflow diagram · grounded in source
1
Developer submits code change
trigger
“When a developer submits a change on Uber's code review platform”
2
File eligibility filtering
validation
“uReview first determines which files are eligible for automated review. It filters out low-signal targets such as configuration files, generated code, and experimental directories.”
3
Context-rich prompt construction
integration
“the system builds a structured prompt that includes surrounding code context such as nearby functions, class definitions, and import statements. This context helps the language model produce precise and relevant suggestions.”
4
Specialized assistant comment generation
ai_action
“uReview uses a pluggable assistant framework, where each assistant focuses on a specific class of issues. This pluggable framework allows each assistant to be developed and evaluated independently, and use customized prompts and context.…”
5
Multi-layered quality filtering
validation
“A secondary prompt evaluates each comment's quality and assigns a confidence score. The prompt is customized for each assistant type, and confidence thresholds for pruning are set at a fine-grained level (per assistant, per language, and…”
6
Comment delivery on review platform
output
“The system posts validated comments directly on the code review platform, in line with the code”
7
Developer feedback and metadata streaming
feedback_loop
“Developers can rate each comment as "Useful" or "Not Useful" and optionally add a note. All comments, along with their associated metadata—including assistant origin, category, confidence score, and developer feedback—are streamed to Apa…”
8
Automated comment verification
feedback_loop
“It automatically evaluates if a given posted comment has been addressed by re-running uReview five times on the final commit”
Reported outcome

uReview is deployed across all six of Uber's monorepos, analyzes over 90% of the approximately 65,000 weekly diffs, maintains a 75% usefulness rate, sees 65% of its comments addressed in the same changeset, and saves approximately 1,500 developer hours per week—equivalent to nearly 39 developer years annually—with a median review turnaround of 4 minutes.

Reported metrics
Weekly diffs analyzed (coverage rate)over 90%
weekly diffs at Uber~65,000
Comment usefulness rate75%
Comments addressed in same changeset65%
Show all 11 reported metrics
weekly diffs analyzed (coverage rate)over 90%
weekly diffs at Uber~65,000
comment usefulness rate75%
comments addressed in same changeset65%
human reviewer comment address rate (benchmark comparison)51%
median CI review turnaround4 minutes
developer hours saved weeklyapproximately 1,500 hours
developer years saved annuallynearly 39 developer years
weekly commits processedover 10,000
human equivalent review time per commit10 minutes per commit
AI cost vs third-party toolsorder of magnitude less
Reported stack
uReviewCommenterFixerClaude-4-Sonneto4-mini-highApache HiveApache KafkaPhabricator
Source
https://www.uber.com/us/en/blog/ureview/
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

uReview is deployed across all six of Uber's monorepos, analyzes over 90% of the approximately 65,000 weekly diffs, maintains a 75% usefulness rate, sees 65% of its comments addressed in the same changeset, and saves…

What tools did this team use?

uReview, Commenter, Fixer, Claude-4-Sonnet, o4-mini-high, Apache Hive, Apache Kafka, Phabricator.

What results were reported?

Weekly diffs analyzed (coverage rate): over 90%; weekly diffs at Uber: ~65,000; Comment usefulness rate: 75%; Comments addressed in same changeset: 65% (source-reported, not independently verified).

What failed first in this deployment?

Third-party AI code-review tools were evaluated and found unsuitable: most required GitHub (Uber uses Phabricator), suffered from many false positives and low-value true positives, and could not interact with Uber's i…

How is this quality assurance AI workflow structured?

Developer submits code change → File eligibility filtering → Context-rich prompt construction → Specialized assistant comment generation → Multi-layered quality filtering → Comment delivery on review platform → Developer feedback and metadata streaming → Automated comment verification.

Related quality assurance cases