Cloudflare Automates Code Review for an Army of Engineers
At Cloudflare, a typical month sees engineers submitting tens of thousands of code changes. The manual review process for these merge requests had become a significant bottleneck, with developers...
At Cloudflare, a typical month sees engineers submitting tens of thousands of code changes. The manual review process for these merge requests had become a significant bottleneck, with developers often waiting hours for feedback. The company's engineering team has fundamentally changed that workflow by deploying an automated, multi-agent AI system directly into their continuous integration pipeline.
Built around the open-source OpenCode agent, the system activates with every GitLab merge request. A coordinator agent deploys up to seven specialized reviewers, each examining a specific area like security, performance, or documentation. This team-in-a-box consolidates its findings into a single, actionable comment posted to the developer. Between March and April, the system processed over 131,000 reviews across 48,000 merge requests, delivering a median result in under four minutes at an average cost of about $1.19.
The design emphasizes practicality and control. Reviews are tiered based on the change's complexity and risk, avoiding over-analysis for minor updates. A plugin architecture allows Cloudflare to swap AI models, source code providers, and internal rule sets. The system uses a mix of models—including Claude Opus and GPT-5.4 for coordination—routed through the company's own AI Gateway for efficiency and significant cost savings. It incorporates a memory layer, allowing it to learn from past interactions and reduce repetitive, irrelevant feedback over time.
Adoption is widespread internally, with the tool now serving over 3,600 users. It has processed hundreds of billions of tokens and handles millions of requests monthly. While the approach echoes industry movements toward AI-assisted development, Cloudflare's implementation is notable for its scale and deep integration with the company's own infrastructure stack. The result is a measurable acceleration in development velocity, letting engineers concentrate on complex problems while automated agents handle routine review checks.
Source: Webpronews
Ready to Modernize Your Business?
Get your AI automation roadmap in minutes, not months.
Analyze Your Workflows →