Anthropic launched Claude Managed Agents, a cloud service designed for building artificial intelligence agents. The service aims to reduce the development workflow from months to weeks.
Deploying a production-grade agent typically requires software teams to build the agent, configure a container, establish infrastructure, and implement observability features. Claude Managed Agents automates many of these tasks and is accessible via application programming interfaces, with customers billed for Claude model usage at eight cents per agent runtime hour.
Developers initiate projects by describing the tasks they wish to automate and specifying tools or third-party applications that the agent should utilize. They can also set cybersecurity rules for tool activation.
Each agent is automatically assigned an isolated container hosting software components defined by the developer. For instance, a website design agent could include a browser within its sandbox. The service significantly simplifies state management, which involves managing the data needed by AI agents to execute tasks. This includes routing programming advice from the web and handling sensitive data such as login credentials.
Tool orchestration is another key feature, enabling the system to determine which tools to deploy for each prompt. An error recovery mechanism allows agents to resume tasks after interruptions.
Two features are currently in research preview. One allows agents to create additional agents for complex tasks, while the other improves the quality of prompt responses automatically. Anthropic stated that the automatic response quality improvement increased outcome task success by up to 10 points during internal testing.
Initial users of Claude Managed Agents include notable tech firms such as Notion Inc., Rakuten Group Inc., and Asana Inc. Several of these companies have already integrated AI agents created with the service into their products.
