Claude Managed Agents lets agents write multi-agent workflows for the server to run, in beta

Anthropic has added dynamic workflows to Claude Managed Agents, its hosted service for running agents. According to the company's release notes, the feature is in beta and the entry is dated October 9, 2026.

The idea is that an agent facing a job with many pieces, such as reviewing hundreds of documents, can write a workflow itself. Anthropic describes a workflow as a program that runs many agents in phases and combines their results. The server runs that program in the background, and the execution is called a workflow run.

What is new is where the coordination happens. In effect, rather than one agent working through a long job step by step, the agent writes a program and the server executes it. Anthropic does not say in this note how the work is divided or how results are merged.

Developers who want to try it need to send the managed-agents-2026-04-01 beta header and set the agent's multiagent field to {"type": "multiagent_20261001", "workflows": {"type": "enabled"}}. Anthropic says developers can steer the agent by telling it in the system prompt when to start a run, and can follow each run through workflow_run. events on the session's event stream.

That is the full extent of what is confirmed here. The release note is short, and it is the only source this article relies on. We found no independent confirmation; the outside write-ups we checked draw on Anthropic's own documentation. The note does not state limits on how many agents a run can use, what a run costs, how failures are reported, or how well workflows perform compared with a single agent. A separate Anthropic page called Workflow runs is referenced but was not used as a source.

Jobs with many pieces, like the document review Anthropic gives as an example, are where a single agent working alone can struggle; that is our analysis, and the note does not say how much the workflows help. Whether they help reliably is not yet shown by independent testing. Developers should read Anthropic's full documentation and test on small jobs before depending on it, since Anthropic labels this a beta.

Revised 16 October 2026: Guth re-read the original sources and rewrote this article. The earlier version remains available as revision 1.