How we’re rethinking work at Cloudflare with Cloudflare OS
- Source
- blog.cloudflare.com
- Author
- Sam Rhea
- Date
Context
Cloudflare changed how it generated a recurring help-desk report: AI wrote reusable application code, so a fresh model session was no longer needed each time the report loaded. Its internal rollout account contrasts this with an earlier workflow that repeatedly spent thousands of recreating largely the same report.
The resulting application queries the ticket data through a service that governs access. Model use remains available for tasks such as drafting replies, which a person can review before sending. The reported benefit comes from separating repeatable computation from the steps that need AI judgment; the claim of zero tokens applies specifically to loading the initial report.
- The tell that AI adoption had outrun governance: a salesperson asked Cloudflare's CIO for production API keys to a dozen systems of record plus deploy-pipeline admin rights, to run an app they had built with AI.
- Cloudflare's rule is that an AI tool never widens your access, and when you share an with a colleague it resolves to that person's permissions, not the permissions of whoever deployed it.
- They wrote an Engineering Codex, an opinionated layer with a named owner for each part of the codebase. Policies say what you can't do; the Codex says what you should do, and agents read it to plan work and review merges and designs.
- Those review agents flagged nearly 250,000 potential problems and blocked 16,000 merges in four months, and caught architectural issues in close to 600 technical designs before a line of code was written.
- Handing non-engineers a coding produced a flood of vibe-coded apps looking for a problem. The fix was a 'magic email' alias quietly staffed by humans using AI: people won't send app ideas to a bot, but will send work they hate doing.
- token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
- agent harness — The scaffolding around a model that turns it into a working agent — the loop, the tools it can call, and the rules for when to stop.
“Anyone at Cloudflare could now write bad code, faster, thanks to AI.”
“If you give everyone a harness workspace that is great at writing code, you’ll wind up with way more code than you need. The result became a flood of vibe coded apps looking for a problem to solve.”
“In the past four months, those agents have flagged nearly a quarter of a million potential problems and blocked 16,000 merges.”
“For some reason, people are less willing to send their vibe coding ideas to what they think is an automated system, but very willing to send the work they do not want to do.”
“In the last month alone, we estimate that our sales team members have saved more than 10,000 hours of time spent on previously manual tasks like territory planning and proposal creation.”
Checking sign-in…
Loading comments…