by Chris DePuy, 650 Group / October 19, 2026
This week, Cloudflare announced an expanded partner ecosystem for Cloudflare OS, the company’s AI workspace. The announcement pairs one-click integrations into widely deployed enterprise applications with services teams that stand the platform up for customers. The press release lists new integrations with Asana, Box, Canva, ClickHouse, GitHub, Gong, and Notion — plus more to come — so administrators can connect systems of record in minutes, and it names EPAM, Happy Cog, and Towards AI as the deployment partners charged with turning how an organization actually works into agents running inside the workspace. Cloudflare OS is basically an open-source internal ChatGPT clone for enterprises has 3 parts: (a) agents are preloaded with company knowledge, (b) there’s a sandbox where agents can build small apps that are shareable, and (c) there are gatekeepers, which is a security layer where each agent starts with limited access and can build up with human approval.
The announcement aims directly at the shadow-AI problem every CIO now manages. Cloudflare’s framing is that employees pick up AI tools without waiting for IT because the tools make them faster, and the side effects accumulate: sensitive files land in AI systems nobody approved, AI-generated code ships without review, and spending climbs with no one metering it. IT is left choosing between locking AI down and slowing the business, or letting it spread without guardrails. “AI tools shouldn’t force companies to choose between moving fast and staying secure,” said Matthew Prince, Cloudflare’s co-founder and CEO, in the announcement.
The mechanics are straightforward. An employee describes the automation or internal tool they need; Cloudflare OS builds it, manages its deployment, and wires it into the systems of record the task touches, without handing over more data access than the job requires. The workspace runs in Cloudflare’s infrastructure rather than on employee laptops, which is what gives security teams the visibility and control that desktop-resident AI tools never offered, and Cloudflare’s AI Gateway routes each job to an appropriate model while keeping token spending visible. Cloudflare describes the platform as an open layer for agents, apps, and work, and the deployment partners describe their role as curating company context and wiring in guardrails so agents behave inside enterprise rules rather than around them.
We read this as the opening move in a control-plane market for enterprise AI, and Cloudflare is entering it from the network side rather than the productivity-suite side. The model layer is commoditizing fast; the broker that connects agents to systems of record — with identity, policy, and cost attached to every connection — is where the durable margin sits. The integration list is the tell. Agents that read Box and ClickHouse, write to Asana and Notion, and act on Gong’s revenue data put Cloudflare on the path of both the inference traffic and the work products, which is the same land-and-expand sequence the company ran against web traffic, now pointed at agentic AI. The cadence backs that read: Cloudflare OS launched as an open-source workspace in early August, built on the platform Cloudflare runs its own global workforce on, and roughly two and a half months later it carries named integrations plus three services partners taking it into enterprises. We are bullish on enterprise AI infrastructure spending, and the workspace layer is one of the few places that monetizes even as model pricing keeps falling.
The obvious competition is the productivity suite itself, since Microsoft and Google ship AI assistants inside the applications most employees already open every morning. Cloudflare’s counter is that useful agents have to reach systems of record across suites and homegrown applications — a cross-suite layer the suite vendors do not broker neutrally — and that an open-source workspace running in the customer’s own Cloudflare account leaves an organization’s processes and connections un-locked from any single vendor. For operators the near-term value is real: least-privilege agent scaffolding in minutes instead of a multi-year custom build. The trade-off is one more control plane to govern, and we would expect consolidation pressure to follow as workspaces, agents, and gateways multiply inside the same enterprise.
650 Group researches several of the technologies we covered in this blog, including Enterprise, Security and SASE, and more.
