What changed
On 9 October 2026, ClaudeDevs said Claude Managed Agents dynamic workflows are available in public beta. The account called this a new type of multiagent orchestration, built for ambitious workloads. In the same post, ClaudeDevs said a lead agent writes a plan that runs across many agents in phases and combines the results at the end. Both ClaudeDevs posts are timestamped 9 October 2026 at 16:12 UTC.
A second ClaudeDevs post from that minute says to configure the agent with multiagent type multiagent_20261001 and then ask Claude to run a workflow. ClaudeDevs said Claude writes the plan and handles orchestration of up to 1,000 agents per run. That how-to post gives the type string and does not name a settings file or a field path for it.
A scoped start, with price and access unstated
ClaudeDevs framed the beta for ambitious workloads. The account said dynamic workflows can use a lot of tokens, and suggested a scoped task first, then a step up in complexity. The thread gives no token price, quota, or measured usage.
Neither post names a price, a plan, a region, an organization rule, or a general-availability date. The agent quickstart page ClaudeDevs linked is a Claude Platform sign-in screen. Its visible line says the platform is for creating agents and applications with frontier Claude models and managed agent infrastructure.
How to try it
ClaudeDevs said to start in Claude Code with /claude-api managed-agents-onboard bug-hunter. bug-hunter is the example ClaudeDevs printed. That post points to the agent quickstart for templates.
The Claude Code changelog lists /claude-api managed-agents-onboard in two Added bullets, and those bullets carry no version heading. One sets up the Managed Agents pattern a page describes as `ant apply` files. The next builds a Console quickstart template, such as deep-researcher, with the ant CLI. deep-researcher is the changelog's example. In the changelog, Managed Agents is the name of the pattern that command sets up. ClaudeDevs's name for the public beta is dynamic workflows: a lead agent writes a plan, runs it across many agents in phases, and combines the results.
The quickstart address ClaudeDevs gave, https://platform.claude.com/agent-quickstart, is a Claude Platform sign-in screen. The line on that screen says the platform is for creating agents and applications with frontier Claude models and managed agent infrastructure. Template steps are not on that screen.
What is still open
Public beta, the type multiagent_20261001, and orchestration of up to 1,000 agents per run are ClaudeDevs's statements from 9 October 2026. The changelog's onboard bullets confirm /claude-api managed-agents-onboard and name deep-researcher. They leave bug-hunter, any token figure, and the agent cap with the ClaudeDevs posts.
Source
Also from
Sources and editorial approach
This article relies on two ClaudeDevs posts from 9 October 2026 at 16:12 UTC, the public-beta announcement and the how-to posted the same minute, plus the Claude Code changelog and the Claude Platform agent quickstart page. Those pages were checked on 9 October 2026. The quickstart page is a sign-in screen, and template steps are not on it. The changelog's onboard bullets have no version heading beside them. These sources do not give a price, a plan, a region, or a general-availability date.