30.8% Faster PRs: How AI-Driven Rovo Dev Code Reviewer Improved the Developer Productivity at Atlassian - Inside Atlassian

Title

30.8% Faster PRs: How AI-Driven Rovo Dev Code Reviewer Improved the Developer Productivity at Atlassian - Inside Atlassian

Content

Atlassian deployed its Rovo Dev Code Reviewer across more than 1,900 internal repositories in a year-long online evaluation, where it cut median pull request cycle time by 30.8 percent and reduced human-written review comments by 35.6 percent, while 38.70 percent of the AI's comments triggered a code change in the subsequent commit compared with 44.45 percent for human comments. Yaotian Zou, Principal Machine Learning Engineer at Atlassian, said the goal was to let engineers focus on high-level logic, architecture, and complex design problems that truly require human expertise, rather than spending hours on repetitive, context-independent parts of code review. Code review has long been a critical pillar of high-quality software engineering at Atlassian, but as products and codebases grew, manual reviews became a significant bottleneck, leading to hours of context switching and delays in shipping features and bug fixes. Rovo Dev Code Reviewer entered open beta before reaching general availability in October 2025, with the accompanying online evaluation research accepted to the 48th IEEE/ACM International Conference on Software Engineering (ICSE 2026) as a collaboration between the Atlassian DevAI and data science teams and Kla Tantithamthavorn of Monash University. Built on Anthropic's Claude 3.5 Sonnet and integrated into Bitbucket, the system runs three stages: review-guided context-aware comment generation, an LLM-as-a-Judge factual correctness check using a cheaper gpt-4o-mini model, and a proprietary resolution-priority classifier trained on more than 50,000 high-quality Rovo Dev-generated comments from internal dogfooding data. Engineers accept or decline suggestions with a single click, and human feedback flows back to refine the agent over time, with readability, bugs, and maintainability emerging as the most resolved comment types. Looking ahead, Atlassian is refining Rovo Dev with advanced context enrichment techniques and expanded agentic capability to support more of the engineering workflow.