Cloudflare

“The week of March 23 hit 10,952, nearly double the Q4 baseline.”

CloudflarePublished by the organization itself (post), no named author

“In the last 30 days, 93% of Cloudflare’s R&D organization used AI coding tools powered by infrastructure we built on our own platform.”

CloudflarePublished by the organization itself (post), no named author

“Over the last 30 days: 100% AI code reviewer coverage across all repos on our standard CI pipeline.”

CloudflarePublished by the organization itself (post), no named author

“The impact on developer velocity internally is clear: we’ve never seen a quarter-to-quarter increase in merge requests to this degree. As AI tooling adoption has grown the 4-week rolling average has climbed from ~5,600/week to over 8,700.”

CloudflarePublished by the organization itself (post), no named author

“we have a security agent that processes over 7 billion tokens per day on Kimi. That would cost an estimated $2.4M per year on a mid-tier proprietary model. But on Workers AI, it’s 77% cheaper.”

CloudflarePublished by the organization itself (post), no named author

“we've been trying to build … platform layer around AI tooling at Cloudflare, not just coding … but the stack that makes the tools useful and safe inside a real engineering organization.”

Rajesh BhatiaSenior Director of Engineering, Cloudflare · Leader

“the code tells the model what's in front of it, but it does not tell the model who owns the service, what the dependencies are, … what databases are being talked to, what conventions are that the teams considers non-negotiable … they reduce the amount of guessing the models have to do. And one thing we learned the hard way is that even with the decent context file, it is better than having none. And when the model stops guessing, the quality goes up everywhere else.”

Rajesh BhatiaSenior Director of Engineering, Cloudflare · Leader

“engineers don't trust AI opinions. They just trust clear findings tied to standards that we … agreed upon. … it stopped sounding more like, maybe this is a problem and started sounding more like this violates a standard that we actually care about.”

Rajesh BhatiaSenior Director of Engineering, Cloudflare · Leader

“sometime in November, we started looking and seeing a lot more benefit for some of the models that we were trying to leverage. I think Opus 4.5 maybe was the inflection point where that then generally made it so that it was very easy to see the kind of results we were seeing and expecting.”

Rajesh BhatiaSenior Director of Engineering, Cloudflare · Leader

“the good part was that we were ready with all the background work that we were doing with respect to the building of the context and building of the right sort of MCP infrastructure internally. So when that became a reality, we started seeing a lot more adoption and because it started helping in the real work that people were doing.”

Rajesh BhatiaSenior Director of Engineering, Cloudflare · Leader

“requests and tokens tell us like usage, like merge requests and pipelines tell us about platform activity, but they do not tell us what value … providing. … the more meaningful long-term question really is like what is the future velocity and outcomes that we can measure with less friction, getting changes out safely.”

Rajesh BhatiaSenior Director of Engineering, Cloudflare · Leader

“Our Network Firewall team audited rampartd using a multi-agent consensus process where every requirement was scored COMPLIANT, PARTIAL, or NON-COMPLIANT with specific violation details and remediation steps reducing what previously required weeks of manual work to a structured, repeatable process.”

CloudflarePublished by the organization itself (post), no named author