Claude Managed Agents: Build a Long-Running Agent in Python

Claude Managed Agents: Build a Long-Running Agent in Python

If you have ever wired up an agent by hand, you know the tax. You write the while-loop, parse tool calls, spawn a sandbox, stream deltas, checkpoint state so a crash does not lose an hour of work, and then babysit all of it in production. Claude Managed Agents is Anthropic's answer to that tax: a hosted agent harness that runs the loop, the sandbox, and the tool execution for you. You define an agent once, point a session at it, and send messages. Claude reads files, runs bash, browses the web, and executes code inside a managed sandbox until the task is done.