Claude Managed Agents is Anthropic's hosted service for building and running Claude-based AI agents. Anthropic announced it on April 8, 2026, as a suite of composable APIs for deploying cloud-hosted agents at scale. Anthropic's launch post describes it as pairing an Anthropic-managed agent harness with infrastructure for state, memory, permissions and scheduled runs.
In practice, you define what an agent should do (its model, instructions, tools and skills), start a session in an execution environment, and stream the events back. Anthropic handles the sandboxing, state persistence across disconnects, authentication and tool execution, so you don't have to build that scaffolding yourself. Governance features include scoped permissions, credential vaults and execution tracing, which you can inspect in the Claude Console.
The status is worth checking, because it changes quickly. Coverage from April described the core service as in public beta with consumption-based pricing, and multi-agent coordination as a research preview. Anthropic's engineering write-up on the design explains the idea of separating the "brain" (the model and its reasoning) from the "hands" (the tools and sandboxes). I haven't checked the current pricing or whether the beta has moved to general availability, so the official docs are the place to confirm those.
If you've built on it or compared it with running your own agent loop, tell us what the tradeoffs looked like for you.
