Long-running async work
Tasks that run for minutes or hours, need multiple tool calls, and benefit from a durable session and managed container.
Use Cases
Claude Managed Agents use cases
Use this page to decide whether your workload actually belongs in Managed Agents before you build around the hosted runtime.
Last updated
2026-04-12
Best Fits
Managed Agents fits best when the runtime itself is part of the product value.
Anthropic’s own overview points to long-running work, managed infrastructure, and stateful sessions as the clearest fit.
Runtime-heavy automation
Workflows that need shell access, files, and a configurable cloud environment without building the sandbox and orchestration layer yourself.
Stateful agent sessions
Apps that need a session you can revisit, steer, interrupt, and inspect later through stored events.
Fit Table
The right choice depends more on runtime ownership than on whether the task sounds “AI.”
Use this table when the product idea sounds agentic but the implementation path is still unclear.
| Dimension | Workload shape | Fit | Why |
|---|---|---|---|
| Thin orchestration | Simple request-response task with a few deterministic tool calls | Weak fit | The Messages API or a smaller app-owned workflow is often enough when you do not need a hosted session runtime. |
| Long-horizon task work | Background code or research task that needs files, shell, and event streaming | Strong fit | This matches the hosted harness, container, and toolset that Managed Agents already provides. |
| Strict infrastructure ownership | Product where execution must stay entirely inside your own network boundary | Case by case | The Agent SDK may be easier if your organization cannot rely on Anthropic-managed runtime boundaries. |
| Interactive developer work | Human operator coding directly in a repo from a terminal or IDE | Poor fit | That is Claude Code territory, not a hosted agent session problem. |
Decision Checklist
Use this checklist before you commit the first sprint.
If you answer no to most of these, Managed Agents is probably not the best starting point.
Would the first version still be valuable if the session runs asynchronously and returns work later?
Do you want Anthropic to provide the harness, container, and event persistence instead of building them?
Can you describe the environment and tool scope in a way that a product review can actually approve?
Are you solving for an app workflow, not for a human developer sitting in Claude Code?
Practical Rule
Start with the workload, not with the most impressive architecture.
This rule of thumb keeps teams from over-building the first release.
A good first project is narrow: one environment, one agent, one clear outcome, and one place where a durable event stream actually helps. If the task only needs a prompt and a response, do not add hosted infrastructure just because it sounds advanced.
Related Pages
Keep moving with the page that matches the next decision.
These pages are the shortest path to the next question most teams hit.
What is Claude Managed Agents?
Use this page when you need a compact explanation of the hosted runtime, the core concepts, and the workloads that match Managed Agents best.
Open pageClaude Managed Agents vs Agent SDK
Use this comparison when the real choice is between Anthropic-managed infrastructure and a programmable SDK stack you operate yourself.
Open pageClaude Managed Agents examples
Use this page when you want examples that match the real managed runtime rather than generic agent demos.
Open pageStay Updated
Want updates as Managed Agents evolves?
Use the waitlist if you want the next update without checking the docs every week.
Waitlist
Want updates as Managed Agents evolves?
Get a short email when this guide adds new examples, API changes, or managed-agents implementation notes.