Case studies / Uber

Minion, uReview, and the internal agent stack

Uber engineers have described Minion as an internal background-agent platform that takes toil (validation-heavy fixes, cleanup, similar interruptions) and returns a pull request for a human. That description comes from talks by Uber staff, not from an uber.com post named Minion. What Uber has published under its own domain is uReview, a GenAI review system that comments on diffs, plus adjacent tools such as FixrLeak. This page keeps those sources separate.

Dated 12 Aug 2025

Scale claims

• Verified uReview analyzes over 90% of the weekly ~65,000 diffs landed at Uber.

  www.uber.com/blog/ureview/ — First-party Engineering blog, "uReview: Scalable, Trustworthy GenAI for Code Review at Uber."

• Verified Engineers who interact with uReview mark 75% of its comments as useful.

  www.uber.com/blog/ureview/ — Same post. This is Uber’s own interaction metric, not an independent audit.

• Verified Over 65% of uReview’s posted comments are addressed.

  www.uber.com/blog/ureview/ — Same post.

• Dropped / flagged Minion generates about 11% of merged pull requests at Uber.

  www.uber.com/blog/ureview/ — Widely repeated from conference talks and recap blogs (Pragmatic Engineer, YouTube titles, background-agents.com). We could not find this figure on uber.com. Dropped.

• Dropped / flagged Minion produces on the order of 1,800 code changes per week.

  www.uber.com/blog/ureview/ — Appears in third-party recaps of Uber talks, not in a first-party post we could open. Dropped.

• Dropped / flagged 95% of Uber engineers use Minion.

  www.uber.com/blog/ureview/ — Newsletter and recap figure. Not on uber.com. Dropped.

Sources

• https://www.uber.com/blog/ureview/
• https://www.uber.com/blog/fixrleak-fixing-java-resource-leaks-with-genai/
• https://www.uber.com/blog/solving-the-agent-identity-crisis/