Claude Managed Agents is Anthropic’s new suite of composable APIs aimed at taking cloud-hosted agents from prototype to production with less bespoke infrastructure work. Announced on April 8, 2026, the launch positions Managed Agents as a public beta feature on the Claude Platform, built around a hosted “agent harness” plus the operational building blocks that typically soak up engineering time.
A hosted agent runtime, not just an agent loop
A recurring pain point in agent development is that the “agent” is rarely just prompts and tools—it’s also the scaffolding required to run safely and reliably in production. Managed Agents bundles those concerns into a managed layer, covering pieces usually implemented per team or per project: sandboxed code execution, checkpointing, credential management, scoped permissions, and end-to-end tracing.
Instead of rewriting orchestration logic as models evolve, Managed Agents provides a built-in harness that decides when to call tools, how to manage context, and how to recover from errors. The goal is to let teams define tasks, tools, and guardrails, while the platform handles the runtime and orchestration mechanics.
Key capabilities Anthropic is emphasizing
Managed Agents’ feature list is framed around production readiness and longer-lived workflows:
- Production-grade agents with secure sandboxing, authentication, and managed tool execution.
- Long-running sessions that can operate autonomously for hours, with progress and outputs persisting through disconnections.
- Multi-agent coordination for parallelizing complex work, where agents can spin up and direct other agents (available as a research preview via an access request form).
- Trusted governance features like scoped permissions, identity management, and execution tracing.
Anthropic also highlights that session tracing, integration analytics, and troubleshooting guidance are integrated into the Claude Console, with the intent that tool calls and failure modes are inspectable during development and operations.
Claude-focused agent iteration (with a research-preview self-eval mode)
Managed Agents is described as being purpose-built for Claude and agentic workloads. Beyond “traditional prompt-and-response,” Anthropic says Managed Agents can support a mode where developers define outcomes and success criteria and Claude self-evaluates and iterates until it gets there—also gated behind a research preview access request.
In internal testing focused on structured file generation, Anthropic reports Managed Agents improved outcome task success “by up to 10 points” compared to a standard prompting loop, with the biggest gains on harder problems.
Early production stories: coding, productivity, and document pipelines
Anthropic points to a range of teams using Managed Agents for production workflows, including coding agents that open PRs, productivity agents that pick up project tasks, and finance/legal agents that extract information from documents. Named examples include:
- Notion, integrating agents in its workspace (private alpha inside Notion Custom Agents).
- Rakuten, deploying specialist agents across multiple internal functions.
- Asana, using Managed Agents to speed up development of AI Teammates.
- Vibecode, using Managed Agents as a default integration to move from prompt to deployed app.
- Sentry, pairing its debugging agent with a Claude-powered agent that writes patches and opens PRs.
Availability and pricing
Managed Agents is available now in public beta on the Claude Platform. Pricing is consumption-based: standard Claude Platform token rates apply, plus $0.08 per session-hour for active runtime, with details in the pricing docs at Claude Managed Agents pricing.
For onboarding, Anthropic links to the Managed Agents overview docs and the Claude Console agent quickstart, and mentions a new CLI for deploying an agent.