Anthropic launches Claude Managed Agents in public beta to simplify and speed up enterprise AI deployments by providing a production-grade agent infrastructure.

The service handles operational complexities such as provisioning, session limits, failures, token usage, rate limits, and long-running workflows so teams can focus on agent logic.

Developers can build agents that operate over hours with controllable tool access and permissions in a secure, scalable setup.

The product targets problems like context forgetting, premature task termination, and difficulties with long-running assignments under context-length limits.

It enables sequential task execution, tool use, code execution, file editing, external-service interactions, and resilience after failures.

Some features, including long-running multi-agent coordination and continual self-evaluation, remain in research preview rather than general availability.

Best use cases include complex, multi-step tasks over extended periods, such as managing workflows, data analysis, and automating processes that require planning and tool use.

Industry context shows growing interest in agentic AI systems, with various players and initiatives shaping the ecosystem.

Initialization requires task descriptions, tool selection, and cybersecurity rules governing tool activation and permissions.

Competition and pricing will be critical; Anthropic must prove reliability, observability, and security at scale, with pricing potentially favoring larger enterprises or aggressive offers to smaller teams.

Key features include secure sandboxed code execution, authentication, checkpointing, scoped permissions, persistent long-running sessions, and session tracing via Claude Console.

The service offers composable APIs, secure sandboxing, authentication, checkpointing, scoped permissions, and persistent sessions to support both simple and multi-agent pipelines, including research-preview coordination and session tracing.