Claude agents can now write their own workflows, splitting hundreds of documents into staged background runs
Synopsis
Claude Managed Agents on the Claude Platform can now write their own workflows in beta for work with many pieces, such as reviewing hundreds of documents. A workflow is a program that runs many agents in phases and combines their results, and the server runs it in the background as a workflow run. It is enabled by setting the agent's multiagent field to multiagent_20261001 with workflows enabled, and each run is followed through workflow_run.* events on the session's event stream.
Interpretation
A Claude Managed Agents agent on the Claude Platform can now write a workflow itself, splitting work with many pieces across several agents in phases. Previously the agent proceeded turn by turn within a session, with no mechanism for the agent to organize multi-phase parallel division of labor for work such as reviewing hundreds of documents. The capability is offered in beta under the managed-agents-2026-04-01 beta header, and the platform names reviewing hundreds of documents as the kind of work with many pieces it targets.
A workflow is a program that runs many agents in phases and combines their results, and the server runs it in the background as a workflow run. It is turned on by setting the agent's multiagent field to {"type": "multiagent_20261001", "workflows": {"type": "enabled"}}, and the system prompt tells the agent when to start a run. Each run is followed through workflow_run.* events on the session's event stream, which is how developers observe the staged agents as they execute.
Perspective
The capability is aimed at developers building applications with Claude Managed Agents on the Claude Platform and is offered in beta, requiring the managed-agents-2026-04-01 beta header. The platform names reviewing hundreds of documents as the kind of work with many pieces it targets; the server runs the workflow in the background and progress is followed through workflow_run.* events on the session's event stream.
The capability remains in beta, and the release notes give no comparison results for workflow runs on quality, latency, or cost. The agent must be told in its system prompt when to start a run, which leaves the timing judgment to the user.