Claude Managed Agents dynamic workflows are now available in public beta. It's a new type of multiagent orchestration, built for your most ambitious workloads. A lead agent writes a plan that runs across many agents in phases, combining the results at the end.

We planted 70 bugs in a 116k-line codebase. Across 3 runs, a single agent found 14, 15 and 27 bugs. A workflow consistently found 66 in each of its 3 runs.

To use it, configure your agent with multiagent type: multiagent_20261001 and then ask Claude to run a workflow. Claude will write the plan and handle orchestration of up to 1,000 agents per run.

Dynamic workflows are powerful and can use a lot of tokens, so we suggest starting with a scoped task. From there, incrementally increase task complexity. Get started in Claude Code with: /claude-api managed-agents-onboard bug-hunter, or check out for templates.

As someone who uses subagents all the time, we'll see how it works

This sounds very Expensive…

what happens when the lead agent's plan is wrong halfway through?

Multiagent orchestration for people who could not manage one agent

That's both exciting and terrifying!

Nice, thank you!

Me: That's a hell of a lot of tokens Also me: gonna try this!!

a lead agent writing a plan for the other agents is just a tiny engineering manager that never schedules a meeting, i respect that

I did this yesterday? 🤣🫣

Productivity increase = x1000. Great now we can move forward much faster, with a model and subagents that are completely reliable. Even so, we'll still need to review the work, but this is a huge step forward!

Outsourcing your orchestration logic to a proprietary model vendor is the ultimate infrastructure trap. Orchestrate in deterministic code; route to open or closed models at the edge.

parallel work is a lot more useful when each task knows exactly what it needs from the others.

the sabi cap packs 100k sensors into one head. this packs many agents under one planner. opposite bets on reasoning at scale – and both shipped without the one metric that would settle it

The main benefit seems to be keeping complex work organized without forcing one agent to handle every piece. Clear handoffs between phases will matter just as much as the agents doing the work.

When reset?

Playing around with managed agents! Pretty sweet so far

Dots are dead already dont bury them

Curious how the lead handles agents reaching conflicting conclusions. Combining their results needs more judgment than collecting a bunch of summaries.

orchestration of up to 1,000 agents per run, i need to test it myself

I described the pampline of the whole idea of using AI very honestly

单agent找bug那活儿都那么能烧，这套多agent编排跑一轮我得先去接两单外包。

what happens when the lead agent's plan is wrong from step 1? does the whole team just confidently build the wrong thing?

So basically, agents now have their own project manager 😂

1000 agents at once 🤯

Claude shipping daily cool

This is nice. Keep them coming Antnropic

So basically, agents now have their own project manager 😂

I have been experimenting with this , architecturally it's sound allows most complex flows, but lead agent is currently not writing right script, efficiency is missing , which could have been done by two sub agents in sequence, it used 70 and still produces stupid artifact