With Dynamic Workers, Sandboxes, Artifacts, and the Think framework, the company aims to help AI agents evolve from experiments on local laptops to full-fledged workloads on the Cloudflare network.
“The way people build software is fundamentally changing. We are entering a world where agents are the ones writing and executing code,” says Matthew Prince, CEO of Cloudflare. “But agents need a home that is secure by default, scales to millions instantly, and persists across long-running tasks.”
Dynamic Workers, Sandboxes, and Artifacts
The new Dynamic Workers system is an isolate-based runtime that executes AI-generated code in a secure environment. Cloudflare claims that Dynamic Workers start up a hundred times faster than traditional containers and incur only a fraction of the cost, scaling to millions of concurrent executions without warm-up. For longer-running tasks, Cloudflare introduces Sandboxes: full Linux environments where agents clone repositories, install Python packages, and build code.
In addition, the company is launching Artifacts, a Git-compatible storage layer that enables developers to create tens of millions of agent repositories. The Think framework within the Agents SDK focuses on persistence: agents use this to support long-running tasks rather than merely responding to individual prompts.
Building on the acquisition of Replicate, which gave Cloudflare access to over 50,000 AI models, the company is expanding its model catalog further. Developers can choose from OpenAI models and open-source alternatives via a single interface. Switching between providers requires changing just one line of code, Cloudflare promises.