[DigitalToday reporter Chi-gyu Hwang] Anthropic has launched a public beta of Claude Managed Agents, expanding into agent infrastructure that lets companies quickly build and deploy cloud-based agents.

On April 8 (local time), The News Stack reported that Anthropic had not directly provided infrastructure to run agents, aside from Claude Code and Cowork.

Claude Managed Agents is a service that lets users define an agent in natural language or a YAML (YAML Ain't Markup Language) file and run it immediately on Anthropic's platform without configuring infrastructure. Anthropic said, "To deploy an agent in a real operating environment, you need sandboxed code execution, checkpointing, credential management, permission scoping and end-to-end tracing." It said, "Infrastructure work over several months comes first before building user-facing features. Managed Agents cuts this process to one-tenth," it added.

Agents can run for hours, and connections to third-party services are handled through MCP servers.

Claude Managed Agents also provides governance features. Anthropic's strategy is to address concerns at companies that have been reluctant to adopt agents by handling permission scoping, identity management and execution tracking at the platform level.

Some features, including advanced memory tooling, multi-agent orchestration, agent self-evaluation and iteration, are available only as a limited research preview.

The pricing applies Anthropic's standard API rates to model token usage, and adds $0.08 per session per hour based on active run time. Idle time while an agent waits for inputs or tools is excluded from charges. Using the web search feature costs an additional $10 per 1,000 searches.