Claude Managed Agents is Anthropic's hosted infrastructure for developing and operating autonomous AI agents at scale. Instead of creating your own agent loop and sandboxes, you use a REST API to have Anthropic run the harness, session logging, and execution environment for you.
Every line in this book refers to Claude Managed Agents, so you always know how they relate to your deployment decisions.
Why do teams pick Claude Managed Agents?
Teams use Claude Managed Agents to deploy AI agents from prototype to production in days, not months. The platform handles the difficult tasks—session lifecycle, retries, sandbox provisioning, and observability—allowing engineers to focus on agent logic and business value.
Claude Managed Agents offers the following key benefits:
- •Faster time to production: The pre-built harness eliminates weeks of infrastructure development.
- •Reliable long-running sessions: Stateful agents can pause, resume, and maintain context for hours or days.
- •Built-in tooling: Code execution, file editing, shell, web browsing, and MCP integrations are all supported out of the box.
- •Observability and audit trails: Session logs and event traces are available in the Claude Console.
- •Scalable runtime: Run multiple simultaneous sessions without managing your own sandbox fleet.
How Claude Managed Agents works (a quick overview)
With Claude Managed Agents, you configure the model, system prompt, tools, and guardrails before starting a session via API. Anthropic creates a sandbox, executes the agent loop, returns events to you, and saves the session state for further examination or restart.
You can use webhooks to trigger actions, wait for human input, and then allow the agent to proceed when it is ready. This approach is appropriate for long-running AI agents that need to pause, respond to external events, or perform scheduled activities.
Claude Managed Agents vs. Agent SDK: Which Should You Use?
Your operational needs will determine whether you use Claude Managed Agents or the Claude Agent SDK.
- •Claude managed the agents: hosted REST API. Anthropic operates the harness and sandbox. Ideal for production AI agents, background jobs, and long-running workflows.
- •Claude Agent SDK: A Python/TypeScript library that runs in your process. Ideal for latency-sensitive, user-facing interactions that require complete control.
Both support code execution, file operations, browsing, and MCP tools, but Managed Agents alleviate the operational effort. If you want to scale Claude agents in production without developing infrastructure, Claude Managed Agents is typically the simplest option.
Real-world applications for Claude Managed Agents.
Organisations employ Claude Managed Agents to automate complex, multi-step procedures that would be unstable with short-lived calls. Common patterns include the following:
1. Customer support triage: Agents examine tickets, check knowledge bases, draft responses, and escalate as needed.
2. Data pipelines: Long-running AI agents extract, clean, and validate data from several systems using retries and logging.
3. Internal tooling: Agents do code reviews, tests, and documentation updates as part of the CI/CD pipeline.
4. Scheduled operations: Nightly reports, inventory checks, and compliance audits are executed as persistent sessions using webhook triggers.
Notion, for example, leverages Claude Managed Agents to orchestrate agent activities and save costs and latency through rapid caching.
Pricing and Cost Considerations
Claude Managed Agents pricing mixes regular model token rates with a runtime charge for active sessions. As of 2026, active runtime costs approximately $0.08 per session hour, with the first 50 hours per day free for all sessions.
Additional fees may include:
- •Model tokens: Billed at the regular Claude API rates for the selected model.
- •Web searches: $10 per 1,000 searches if the agent conducts a web search within the session.
- •Self-hosted sandboxes: Optional for tighter data residency or cost control, requiring your own compute.
This strategy makes it cost-effective to run multiple parallel production AI agents without overprovisioning infrastructure.
How to deploy Claude Managed Agents (Step by Step)
To put Claude Managed Agents into production, follow this high-level flow:
1. Configure your Claude API access and activate the Managed Agents beta headers.
2. Define your agent: Select a model, system prompt, permitted tools, and guardrails.
3. Set up sandboxes: Use Anthropic-hosted containers or link self-hosted workers/MCP.
4. Begin a session: Use the Managed Agents API to establish a session and stream events.
5. Configure webhooks: Use session.status_idle and session. Budget_reached events for human-in-the-loop and cost control.
6. Monitor and iterate: Examine session records in the Claude Console to improve prompts, tools, and budgets.
This approach allows you to easily install Claude Managed Agents while maintaining complete control over behaviour and expenses.
Best methods for scaling Claude agents in production.
To scale Claude agents in production safely and reliably, use the following practices:
- •Use budgets and timeouts: To avoid unexpected costs, set session budgets and idle timeouts.
- •Use webhooks: Don't keep long-lived HTTP connections; instead, let sessions idle and resume via events.
- •Pin geography as needed: Use inference geo pinning to meet data residency criteria.
- •Begin small, then scale: Test one workflow, measure latency, cost, and quality, and then repeat patterns.
- •Document agent roles: Treat each agent as a microservice with distinct tasks and SLAs.
These procedures assist you in developing a reliable, maintainable managed AI agent infrastructure.
Who Should Use Claude Managed Agents?
Claude Managed Agents is ideal for:
- •Engineering teams are releasing production AI agents but do not wish to run their own sandbox infrastructure.
- •Product teams want long-running AI agents to perform background chores, scheduled jobs, or multi-step workflows.
- •Organisations that prioritise observability, auditability, and a clear separation of agent logic and infrastructure.
If you're deciding between Claude Managed Agents and the Agent SDK, and your priority is operational simplicity and scale, Managed Agents is usually the better option.
Final thoughts.
Claude Managed Agents provides teams with a rapid, dependable path to production AI agents that requires considerably less plumbing. By delegating the harness, sandbox, and session management to Anthropic, you can concentrate on what's important: usable, safe, and valuable agent activity for your users.