Cloudflare has unveiled plans to expand 'Agent Cloud,' which supports the deployment and operation of AI agents. By introducing execution environments, storage capabilities, and tools to support long-term tasks, the company has focused on enabling agents, previously confined to the experimental phase, to be operated in actual service environments.

Cloudflare officially announced the details on April 13. The company stated that it would support developers in building and deploying agents without having to configure separate environments individually. The announcement included features related to Dynamic Workers, Artifacts, Sandbox, and the Agent Development Kit.

Previously, operating agents often relied on always-on servers or containers, which raised issues regarding cost and scalability. Cloudflare explained that to address this, they introduced Dynamic Workers, which are structured to run when needed and clean up after the task is completed.

Code storage and management features were also presented. The company stated that it enables the storage and utilization of code and data generated by agents through Artifacts, a storage structure compatible with Git. It also highlighted that the system is accessible via standard Git clients.

In terms of the execution environment, an isolated Linux-based sandbox was included. Cloudflare stated that in this environment, agents can handle tasks throughout the development process, such as cloning repositories, installing packages, or performing builds.

The agent development kit also includes features to support long-running tasks. The company explained that this enables the handling of multi-step workflows, going beyond simply responding to a single request.

Along with this, a structure was presented that allows for the selective use of various AI models. Cloudflare announced that it supports the utilization of multiple models, including OpenAI models and open-source models, on its platform. OpenAI also introduced that its models and Codexes can be utilized within the Cloudflare environment.

This announcement can be summarized as a comprehensive presentation of the execution, storage, and development environments required for operating AI agents. As the scope of agent utilization expands across the industry, competition for related infrastructure is also expected to continue.