Cloudflare on Tuesday announced an expansion of its Agent Cloud platform, introducing new tools and infrastructure aimed at enabling developers to build and deploy large-scale, autonomous AI agents.
The company said the move is intended to support a shift in software development, where AI agents evolve beyond chatbots to systems capable of executing multi-step tasks, generating code, and operating across workflows.
Cloudflare’s expanded offering includes infrastructure, security, and developer tools designed to move AI agents from experimental environments to production-grade deployments across its global network.
Matthew Prince, co-founder and CEO of Cloudflare, said the company is positioning its platform to support what he described as an emerging “agentic web”, where agents can write and execute code while operating in scalable and secure environments.
The platform introduces features such as Dynamic Workers, an execution model designed to run AI-generated code in isolated environments, and Artifacts, a Git-compatible storage system aimed at supporting large-scale code generation and versioning by AI systems.
The company also announced general availability of Sandboxes, which provide persistent Linux-based environments for agents to perform tasks such as code execution, package installation, and iterative development.
In addition, a new framework called Think has been introduced within the Agents SDK to support long-running and multi-step agent workflows, addressing limitations of short-lived AI processes.
Cloudflare said it is also expanding its AI model ecosystem following its acquisition of Replicate, allowing developers to access multiple proprietary and open-source models, including those from OpenAI, through a unified platform.
The company added that the platform is designed to offer flexibility in model selection, enabling developers to switch between providers with minimal changes to code.
(Source: Company statement)