How Cloudflare Is Rethinking Work with Cloudflare OS
Original title:How we’re rethinking work at Cloudflare with Cloudflare OS
AI Summary
Cloudflare CIO Sam Rhea describes how the company moved from cautious AI experiments in 2025 toward a broader operating model called Cloudflare OS. The shift followed rapid improvements in models and harnesses that made agents capable of performing useful work, not merely answering questions. Hundreds of employees across technical and non-technical teams experimented with new building tools around the New Year. One sales employee’s AI-built “SuperApp,” however, required production access to roughly a dozen systems of record and administrative permissions for a deployment pipeline, highlighting the governance challenge created by bottom-up agent development.
Why it's worth reading
As enterprise agents move from chat interfaces to systems that can act, Cloudflare’s experience offers a timely case study in balancing bottom-up innovation with production access, deployment controls, and organizational responsibility.
Deep Read
1. What happened
Original facts: Cloudflare CIO Sam Rhea says the company recognized roughly six months earlier that internal AI use was exceeding its existing governance model. A sales employee used AI to build an internal application called “SuperApp” and requested production access to about a dozen systems of record, plus administrator permissions for a deployment pipeline. Cloudflare then began rethinking work through Cloudflare OS.\n\n## 2. Core technology Original facts: During 2025, Cloudflare mainly deployed informational chat applications and experimented with AI-generated boilerplate code. By the end of that year, better models and more powerful harnesses made AI agents capable of doing tasks rather than only generating responses.\n\nAnalysis: The important shift is not identified as one particular model. It is the combination of models, harnesses, and permissions to enterprise systems that turns an agent into an active participant in a workflow.\n\n## 3. Key evidence and numbers Original facts: The change in direction occurred at the end of 2025. During the quieter weeks around New Year, hundreds of Cloudflare employees in technical and non-technical roles experimented with easier application-building tools. The SuperApp request involved production access to approximately twelve systems and administrator access to a deployment pipeline.\n\nUnverified inference: These figures suggest that demand for internal AI applications had spread beyond a small pilot group, but the supplied abstract gives no application count, time savings, incident rate, or production-deployment success rate.\n\n## 4. Why it matters Analysis: The enterprise AI bottleneck is shifting from whether a model can answer questions to who is authorized to let an agent act on behalf of the organization. When non-engineering teams can rapidly assemble cross-system applications, innovation speed collides with least privilege, auditability, and release accountability.\n\n## 5. Practical impact Analysis: Enterprises may need to treat agent authorization, systems-of-record boundaries, deployment approvals, credential management, and ownership as one operating-platform problem. Cloudflare’s example also suggests that IT may evolve from centralized tool procurement toward providing controlled build environments and secure paths to production.\n\n## 6. Limitations and uncertainty The abstract does not describe Cloudflare OS’s full architecture, product components, authorization model, scope of adoption, or outcome metrics. It also does not say whether the SuperApp reached production. The source therefore does not establish that the approach reduced costs, improved sales productivity, or eliminated security risk.\n\n## 7. Original sources Cloudflare Blog: How we’re rethinking work at Cloudflare with Cloudflare OS\n\nMetadata supplied with the item: published 2026-08-05; the abstract identifies Sam Rhea as Cloudflare’s Chief Information Officer.