Cloudflare pushes Agent Cloud
Cloudflare expanded its Agent Cloud and framed agents as deployable internet workloads with edge distribution, managed runtimes, networking and security primitives rather than notebook experiments. The announcement and its Agents Week messaging emphasize durable execution, state handling, environment‑level security and observability that tie agent behavior to infrastructure events. The company suggests agent deployment should feel familiar to service engineers—packageable, versioned and observable like other APIs and services. (blog.cloudflare.com) (investing.com)
Cloudflare said Monday, April 13, that it is expanding Agent Cloud so developers can run artificial intelligence agents as production services on its network, not just as local demos. (markets.ft.com) An agent is software that can remember context, call tools, schedule work, and keep working after a user leaves. Cloudflare’s Agents documentation says each agent runs on a Durable Object, a stateful micro-server with its own Structured Query Language database, WebSocket connections, and scheduling. (developers.cloudflare.com) Cloudflare said the new push is meant to move agents from “experimental demos on local laptops” to “production-grade workloads” on its global network. The company announcement says Dynamic Workers spin up in milliseconds, run JavaScript in an isolate-based sandbox, and shut down when the code is done. (markets.ft.com) Most artificial intelligence apps today are stateless, which means they answer one request and forget it. Cloudflare’s pitch is that useful agents need memory, timers, live connections, and a place to store state across restarts, deploys, and hibernation. (developers.cloudflare.com) That design leans on products Cloudflare has been building for more than a year. Its Workflows service became generally available on April 7, 2025, and Cloudflare describes it as a durable execution engine that retries failed steps and resumes long-running jobs from the last saved step. (blog.cloudflare.com) Cloudflare has also been adding the plumbing around those agents. On April 7, 2025, the company said its Agents Software Development Kit added remote Model Context Protocol clients, authentication integrations with Stytch, Auth0, and WorkOS, and hibernation for inactive stateful servers. (blog.cloudflare.com) The observability piece is unusually explicit. Cloudflare’s Agents documentation says the platform emits structured events for remote procedure calls, state updates, schedule execution, workflow transitions, Model Context Protocol connections, and email processing through named diagnostics channels. (developers.cloudflare.com) Cloudflare is also tying the platform more closely to OpenAI. OpenAI said Monday that enterprises can deploy agents built on Codex harness to Cloudflare, that GPT-5.4 is being expanded across Agent Cloud, and that Codex harness is now generally available in Cloudflare Sandboxes. (openai.com) Matthew Prince, Cloudflare’s co-founder and chief executive officer, said in the company announcement that “agents need a home that is secure by default, scales to millions instantly, and persists across long-running tasks.” Monday’s rollout puts that claim at the center of Cloudflare’s case that agents should be deployed like any other internet service. (markets.ft.com)
