Claude Managed Agents dynamic workflows (beta): what Anthropic shipped on Oct 9, 2026
Since Oct 9, 2026, Claude Managed Agents can write workflows that run many agents in phases (beta). How to turn it on, limits, and what it costs.
On Oct 9, 2026, Anthropic let Claude Managed Agents agents write dynamic workflows: programs that run many agents in phases and combine their results. The server runs each one in the background as a workflow run. You turn it on with one multiagent setting and steer it from the system prompt. A run has no price of its own: you pay token rates plus Managed Agents' $0.08 per session-hour runtime. Checked against Anthropic's docs on 2026-10-12.
BETA. Dynamic workflows ship in beta behind the
managed-agents-2026-04-01beta header, which all Managed Agents endpoints require (the SDK sets it automatically). Anthropic says "Behaviors may be refined between releases to improve outputs." A general-availability date is not published as of 2026-10-12.
What Anthropic announced
From the Claude Platform release notes, Oct 9, 2026:
"You can now let a Claude Managed Agents agent use dynamic workflows, in beta with the
managed-agents-2026-04-01beta header. For work with many pieces, such as reviewing hundreds of documents, the agent can write a workflow. A workflow is a program that runs many agents in phases and combines their results. The server runs it in the background as a workflow run. To turn dynamic workflows on, set the agent'smultiagentfield to{"type": "multiagent_20261001", "workflows": {"type": "enabled"}}. To guide the agent, tell it in its system prompt when to start a run. Follow each run throughworkflow_run.*events on the session's event stream."
Managed Agents access is enabled by default for all API accounts.
How a workflow run works
The agent decides when to start a run; there's no extra API call for you to make. The run executes in the background while the agent keeps working or ends its turn. Context and results pass from one agent to the next programmatically, which leaves the main session thread free to talk to the user and report on progress.
A run can use two kinds of agents:
- Inline agents, which the workflow defines itself. They use the session agent's model and a subset of its tools.
- Predefined agents, which you list in
workflows.predefined_agents(up to 20). That list is separate fromsubagents.predefined_agents.
Turn it on
Set multiagent.type to "multiagent_20261001" with workflows enabled. Watch out: this type turns on both dynamic workflows and subagents by default. For workflows only, add "subagents": {"type": "disabled"}.
Then tell the agent when a run is worth it. Anthropic's example system prompt:
"You review contracts. When you're asked to review more than a few contracts, start a workflow run that reads them in parallel and combines the findings. Review one or two contracts yourself, without a run."
The Claude moves worth stealing, by email
Workflows tested on real jobs, sent only when they are worth it. Free.
Follow a run
Each run reports on the session's event stream:
| Event | Meaning |
|---|---|
workflow_run.created | Run started, with its phases |
workflow_run.status_running / status_idle | Working, or paused (for example at the session budget) |
workflow_run.phase_started / phase_ended | A phase opened or closed |
workflow_run.status_ended | Run finished, with a result |
workflow_run.error | Something failed, such as too many open runs |
Limits
Anthropic says it doesn't guarantee these numbers, so they can change:
| Limit | Value |
|---|---|
| Threads working at once in one run | 64 |
| Agents a workflow starts over the run's life | 1,000 (then thread_limit_error) |
| Run lifetime | 24 hours by default (then timeout_error) |
| Runs open at once in a session | 10 by default (then max_workflow_runs_error) |
What it costs
The workflow-runs doc is direct: "A run has no price of its own. The tokens its agents use are billed like the session's other tokens, at each model's rates."
So the normal Managed Agents bill applies: tokens at standard model pricing (prompt caching multipliers apply, US-only inference is 1.1x, no Batch discount), plus session runtime at $0.08 per session-hour, counted only while the session is running. A run's model requests count toward the session budget, and every open run pauses when that budget is hit. See our Claude pricing guide for model rates.
Workflows vs subagents
A subagent's thread stays available, so the agent can send it follow-ups. A run's threads can't receive follow-ups, and the server archives each one by the end of the run.
These are Platform API features. Claude Code's CLI has its own subagents and agent teams, which work differently.
FAQ
Is it generally available?
No. It's beta, behind the managed-agents-2026-04-01 header.
Does a workflow run cost extra? No separate price. You pay for tokens and session runtime.
Is it on Amazon Bedrock or Google Cloud? Anthropic's pricing doc says Managed Agents is "not available on partner-operated cloud platforms". It is on Claude Platform on AWS "with some differences in feature availability"; whether dynamic workflows are included there is not published as of 2026-10-12.
Sources (checked 2026-10-12 08:10 CEST)
- Claude Platform release notes, Oct 9, 2026: https://platform.claude.com/docs/en/release-notes/overview
- Multiagent orchestration, Dynamic workflows: https://platform.claude.com/docs/en/managed-agents/multiagent-orchestration#dynamic-workflows
- Workflow runs: https://platform.claude.com/docs/en/managed-agents/workflow-runs
- Claude Managed Agents overview: https://platform.claude.com/docs/en/managed-agents/overview
- Pricing, Claude Managed Agents pricing: https://platform.claude.com/docs/en/about-claude/pricing#claude-managed-agents-pricing