The Evolution of the Agentic Surface: Building with Claude Managed Agents

Beyond simple prompt input, applying AI agents to real-world services requires code execution environments, data access permissions, scalable infrastructure, and observability. Previously, developers had to build the agent's loop and infrastructure themselves, which caused complex issues such as security, state management, and permission configuration.

Claude Managed Agents is a set of composable APIs for building production-grade agents. This service is built on the proven harness of Claude Code, and helps developers shorten the time from prototype to actual service launch from months to days.

The key challenges addressed are as follows:

• Hosting and Scaling: Maintaining processes for long-running tasks and providing scalability according to usage
• Session Management: The ability to store the agent's task history and progress state, and resume after interruption
• File System Management: Providing an isolated workspace where the agent can edit code and generate outputs
• Execution Isolation and Security: An isolated environment where agent-generated code can run safely, and credentials management
• Observability: The ability to reconstruct and trace every step performed by the agent