Anthropic PBC has introduced Claude Managed Agents, a cloud service aimed at significantly reducing the time required to build artificial intelligence agents. The company states this new offering can shorten the development workflow from months down to mere weeks.
Developing production-grade AI agents typically involves extensive scaffolding beyond the agent itself. Software teams must configure secure containers for agent tasks, set up underlying infrastructure, and implement observability features. Claude Managed Agents automates many of these complex processes through a set of application programming interfaces.
Customers will be billed based on Claude model usage and a runtime fee of $0.08 per agent hour. Developers initiate a project by describing desired automation tasks and specifying required third-party tools. Cybersecurity rules can also be defined to control agent actions, such as requiring user permission before activating a tool.
The service automatically provisions isolated containers for each agent, hosting pre-defined software components. This includes features for state management, crucial for handling data agents use to complete tasks, ranging from web information to sensitive login credentials.
Claude Managed Agents also simplifies tool orchestration, determining the appropriate tools for generating responses to agent prompts. An error recovery mechanism ensures work can resume seamlessly after an outage.
Currently in research preview, two key features allow agents to spin up other agents for complex tasks and automatically refine prompt response quality, which has shown significant improvements in internal testing.