ACCORDING to The Cloudflare Blog, Cloudflare has built a CI-native orchestration system around OpenCode to run AI-assisted code reviews at scale, with up to seven specialised reviewers handling security, performance, code quality, documentation, release management, and compliance.
The internal rollout has processed tens of thousands of merge requests, with the first month delivering 131,246 review runs across 48,095 merge requests in 5,169 repositories, and an average MR being reviewed 2.7 times in around 3 minutes 39 seconds per review. The system uses a coordinator factor to deduplicate findings and post a single structured review comment, and it can block merges for any critical issue or security vulnerability.
A key design choice is a plugin-based architecture enabling GitLab and future providers, plus a shared OpenCode workflow that reads per-file patches and a shared context file to cut token costs. Cloudflare reports a strong cost and token-usage story, with a cache hit rate of 85.7% and about 120 billion tokens processed in the period, highlighting the balance between speed and cost in automated code review.