Cloudflare targets production-grade AI agents
Cloudflare has expanded its Agent Cloud platform with a set of new tools aimed at moving AI agents from development experiments to production-grade enterprise workloads.
The expansion includes Dynamic Workers, a new isolate-based runtime for running AI-generated code in sandboxed environments. Cloudflare says Dynamic Workers spin up in milliseconds and cost significantly less than traditional containers. The company claims they run at 100 times the speed of containers at a fraction of the cost.
Cloudflare is also introducing Artifacts, a Git-compatible storage layer built for AI-generated code. The company says Artifacts allows developers to create tens of millions of repositories and fork from remote sources, giving agents persistent storage accessible through standard Git clients.
Sandboxes, a persistent Linux environment for agents requiring a full operating system, has moved to general availability. A sandbox provides a shell, filesystem, and background processes, enabling agents to install packages, run builds, and iterate across tasks.
A new SDK framework called Think is designed to support long-running, multi-step agent tasks, addressing what Cloudflare describes as the limitation of agents that can only respond to single prompts without persisting state across sessions.
Cloudflare is also expanding its AI model catalogue, following its acquisition of Replicate, to include proprietary models from OpenAI and open-source alternatives through a single interface. The company says developers can switch between model providers by changing a single line of code.
"The way people build software is fundamentally changing. We are entering a world where agents are the ones writing and executing code," said Matthew Prince, co-founder and CEO of Cloudflare. "But agents need a home that is secure by default, scales to millions instantly, and persists across long-running tasks."
OpenAI's Rohan Varma, who leads product for Codex, cited the partnership in the announcement, stating that the integration makes it easier to deploy production-ready agents using GPT-5.4 and Codex for enterprise workloads.