How Cloudflare enforces engineering standards using AI

Cloudflare developed the Cloudflare Codex, a governed set of engineering standards structured as RFCs with MUST/SHOULD keywords, and made them machine-readable by extracting requirements into JSON. AI agents now consume these standards to enforce consistency across code reviews, spec reviews, and incident report reviews. Over four months, the AI code reviewer flagged close to 230,000 deviations and withheld or blocked 16,000 merges when enforced MUST requirements were violated. The spec reviewer evaluated nearly 600 technical designs before implementation, catching architectural mistakes early. An incident report reviewer assessed more than 200 postmortems for completeness and follow-up actions. By codifying institutional knowledge and automating enforcement, Cloudflare catches issues sooner, reduces drift between teams, and lets engineers focus on higher-level decisions instead of searching for scattered guidance. The system is being expanded beyond engineering to product, security, and compliance teams.