Cloudflare has announced an expansion of its Agent Cloud, introducing new features to support developers in building, deploying, and scaling AI agents across its global network.
These updates provide infrastructure and tools intended to help agents move from experimental stages to production-scale workloads.
Go deeper with GlobalData
Access deeper industry intelligence
Experience unmatched clarity with a single platform that combines unique data, AI, and human expertise.
The company is responding to a shift within the AI industry, which has moved beyond initial chatbot applications towards autonomous coding agents and context-aware tools designed for multi-step operations.
Traditional infrastructure, which typically relies on costly, persistent virtual servers or isolated sandboxes, has been unable to efficiently support the simultaneous operation of many personal agents for each user.
Cloudflare’s Agent Cloud aims to overcome these limitations by handling compute, deployment, and security tasks, allowing developers to focus on building the application.
A key focus of the update is the launch of Dynamic Workers, a new compute model that uses an isolate-based runtime.
This model executes AI-generated code snippets rapidly and securely without the expense of containerisation.
The Dynamic Workers deploy in milliseconds to carry out tasks such as calling APIs, transforming data, and chaining tool operations, before shutting down quickly.
Cloudflare states this approach offers secure isolation, increased speed, and lower costs, and can handle millions of concurrent executions.
The company has introduced Artifacts, a storage system compatible with Git, designed to address the needs of AI agents generating significant volumes of code.
Artifacts enables developers to create large numbers of repositories, fork code from various sources, and store both code and data for agent access through standard Git clients.
For situations where agents require a complete operating system, Cloudflare is making its Sandboxes feature generally available.
Sandboxes provide a persistent, isolated Linux environment for agents to perform functions such as cloning repositories, installing packages, running builds, and managing background processes.
Cloudflare is also adding Think, a framework in the Agent Cloud SDK, to support long-running, persistent agents capable of managing multi-step tasks.
Think is designed for agents that require extended operation and task continuity, addressing a gap left by short-lived application agents.
Following its acquisition of Replicate, Cloudflare is also expanding its model catalogue, enabling developers to access proprietary AI models like those from OpenAI and GPT-5.4, as well as open-source alternatives.
This catalogue is available through a unified interface, permitting developers to switch between model providers by altering a single line of code.
These advancements reflect changes in software development and Cloudflare’s attempt to provide infrastructure capable of supporting large-scale, autonomous, persistent agents.
