Anthropic launched Claude Managed Agents, a cloud service designed for artificial intelligence agent development that reduces workflow duration from months to weeks. The service automates the setup and management processes typically required for deploying operational AI agents.
Development projects necessitate the configuration of containers, infrastructure, and observability features. Claude Managed Agents simplifies these tasks and is accessible via application programming interfaces (APIs). Customers will incur charges based on Claude model usage, plus an hourly fee of eight cents per agent runtime.
To initiate a project, developers provide a description of tasks for automation. They must identify tools or third-party applications for the agent and establish cybersecurity parameters governing tool activation protocols. Each agent operates within an isolated container hosting pre-defined software components. For instance, a web design agent could include a browser within its sandboxed environment.
Claude Managed Agents additionally automates state management tasks, including handling sensitive data like login credentials. It streamlines tool orchestration, selecting the appropriate tools for generating responses to prompts. Moreover, the service features an error recovery mechanism, allowing agents to resume operations post-disruption.
Currently, two features are in research preview: the capability for an agent to create additional agents for complex tasks and a function to enhance prompt response quality. According to Anthropic, this latter feature yielded a 10-point improvement in task success rates during internal testing.
Initial users of Claude Managed Agents include Notion Inc., Rakuten Group Inc., and Asana Inc. Anthropic reports that several customers have successfully integrated AI agents developed through the service into their existing products.
