Releases·Tool Release
Uber engineers detail uReview, their multi-agent code review system for Phabricator monorepos
A YouTube talk by Uber's Will Bond and Ameya Ketkar describes uReview, a multi-agent code review engine Uber built in-house. Code review wait times grew from about three hours in 2024 to nine hours in 2026 as volume and PR size grew, making review the bottleneck for thousands of engineers. Uber chose to build rather than buy because most vendors don't support Phabricator and because it wanted agents held to the same review rules as humans.
23 engagement1 sourceFri, Aug 28, 2026, 02:30 PM
AI Overview
- Code review became the bottleneck for thousands of engineers across hundreds of teams, twelve sites, and six language-specific monorepos
- First-review wait times grew from roughly three hours per PR in 2024 to nine hours in 2026 as volume and PR size grew
- Uber built rather than bought partly because most vendors do not support Phabricator
- Agents run in the inner loop and are reviewed against exactly the same rules as human reviewers
- The talk emphasizes what the team had to measure to make the system work
Related
Developments·Paper
Study analyzes developer responses to AI-generated code review comments across five agents
A large-scale empirical study examined 54,791 code review comments generated by five AI coding agents (Copilot, Cursor, Codex, Devin, Claude) across 342 Python repositories on GitHub. The research investigates how developers resolve agent-generated feedback, revealing patterns in acceptance and rejection of automated review suggestions.
A large-scale empirical study examined 54,791 code review comments generated by five AI coding agents (Copilot, Cursor, Codex, Devin, Claude) across 342 Python repositories on GitHub. The research investigates how developers resolve agent-generated feedback, revealing patterns in acceptance and rejection of automated review suggestions.
0 engagement1 sourcearxivFri, Jul 24, 2026, 05:51 AM
Developments·Paper
Arxiv paper analyzes agentic code reviewer behavior, cost, and human-alignment in terminal
A new arxiv paper presents a trajectory-level analysis of agentic code reviewers in terminal-based environments, finding they achieve higher review precision but incur substantial exploration and validation costs. The study moves beyond performance-centric evaluations to capture dynamic behaviors and hidden operational costs.
A new arxiv paper presents a trajectory-level analysis of agentic code reviewers in terminal-based environments, finding they achieve higher review precision but incur substantial exploration and validation costs. The study moves beyond performance-centric evaluations to capture dynamic behaviors and hidden operational costs.
0 engagement1 sourcearxivSat, Jul 18, 2026, 10:03 AM
Products·building
Bubo: AI code reviewer that learns from human feedback
Bubo is an AI code-review tool that flags issues in pull requests and learns from developers' comments on those findings, reducing noise and false positives over time. It uses LLMs to generate evidence-backed findings or LGTM verdicts, and incorporates human corrections to adapt to each repository's context, including tribal knowledge and business logic.
Bubo is an AI code-review tool that flags issues in pull requests and learns from developers' comments on those findings, reducing noise and false positives over time. It uses LLMs to generate evidence-backed findings or LGTM verdicts, and incorporates human corrections to adapt to each repository's context, including tribal knowledge and business logic.
4 engagement2 sourcesredditTue, Aug 4, 2026, 07:11 AM
Social·Community
AI-authored PRs merge at a third of the rate of human ones, benchmark finds
A benchmark of 8.1 million pull requests across 4,800 teams in 42 countries found that AI-authored PRs wait over 16 hours for review, versus about 200 minutes for human-authored ones, and merge at only a third of the rate. The findings highlight downstream friction for agent-generated code.
A benchmark of 8.1 million pull requests across 4,800 teams in 42 countries found that AI-authored PRs wait over 16 hours for review, versus about 200 minutes for human-authored ones, and merge at only a third of the rate. The findings highlight downstream friction for agent-generated code.
4 engagement1 sourceredditSun, Aug 23, 2026, 09:15 AM
Products·launched
Merge: AI-native code review assessments for engineering hiring
Merge is a platform that assesses engineering candidates through code review tasks instead of traditional coding output. It uses AI to generate or simulate pull requests with embedded bugs, testing candidates' ability to catch issues and evaluate architectural decisions. It addresses the shift towards code review in modern engineering workflows, where AI-generated code is prevalent.
Merge is a platform that assesses engineering candidates through code review tasks instead of traditional coding output. It uses AI to generate or simulate pull requests with embedded bugs, testing candidates' ability to catch issues and evaluate architectural decisions. It addresses the shift towards code review in modern engineering workflows, where AI-generated code is prevalent.
7 engagement2 sourceshackernews, redditFri, Aug 7, 2026, 05:20 PM