AI Judge changed title after evaluation, original title: "Anthropic adds dynamic workflows to Claude Code to automatically spin up and coordinate hundreds of parallel subagents"
Parallel execution can significantly increase token consumption
Danielle Fong 🔆#756
New in Claude Code (research preview): dynamic workflows.
Claude writes an orchestration script on the fly, then spins up a large fleet of coordinated subagents in parallel to take on your most complex tasks.
Use the word "workflow" in a prompt to get started.
Many users praised Claude Code's dynamic workflows for orchestrating hundreds of coordinated AI agents and boosting output dramatically, while others called the feature overhyped, costly, or not innovative.
No Digg Deeper questions have been answered for this story yet.
Excited to share our most powerful new Claude Code feature: dynamic workflows!
Mention "workflow" in a prompt and Claude will dynamically create an orchestration plan that it strictly follows, allowing you to confidently trust that every stage happens in the right order even across 100s of agents.
In case you're curious about why dynamic workflows are so powerful and the future, read the RLM paper! Opus 4.8 + dynamic workflows in Claude Code is perhaps the first instance of a frontier model seriously trained to be an RLM.
I suspect within a year they'll just become the standard for nearly all coding agent interactions.
New in Claude Code (research preview): dynamic workflows.
Claude writes an orchestration script on the fly, then spins up a large fleet of coordinated subagents in parallel to take on your most complex tasks.
Use the word "workflow" in a prompt to get started.
Opus 4.8 is awesome but that's expected. The unsung hero of this release for me is dynamic workflows.
Claude plans your task, fans it out to tens or hundreds of parallel subagents, verifies their work, and iterates until the results converge into one coordinated answer.
Claude code's new dynamic workflows update is absurd.
Make sure you understand what its doing here. This isn't simply a long running mode like /goal, or a fancy subagent verifier process.
This is Claude vibecoding an entire brand new subagent fleet harness on demand
RLM on agent harnesses. Thats what "dynamic workflow" means.
This is basically a new scaling law dimension.
Base Model Compute x Inference time Thinking Compute x Inference time generated Harness Compute
HUGE step forward on the path of AI are you paying attention yet?
New in Claude Code (research preview): dynamic workflows.
Claude writes an orchestration script on the fly, then spins up a large fleet of coordinated subagents in parallel to take on your most complex tasks.
Use the word "workflow" in a prompt to get started.
This is one of the most useful and incredible things we've shipped. @bcherny once called it the future. Give it a whirl!!
New in Claude Code (research preview): dynamic workflows.
Claude writes an orchestration script on the fly, then spins up a large fleet of coordinated subagents in parallel to take on your most complex tasks.
Use the word "workflow" in a prompt to get started.
Excited to share our most powerful new Claude Code feature: dynamic workflows!
Mention "workflow" in a prompt and Claude will dynamically create an orchestration plan that it strictly follows, allowing you to confidently trust that every stage happens in the right order even across 100s of agents.
if you are interested in taking a sneak peek at what might be going on in claude code dynamic workflow feature check out this 2h classic
New in Claude Code (research preview): dynamic workflows.
Claude writes an orchestration script on the fly, then spins up a large fleet of coordinated subagents in parallel to take on your most complex tasks.
Use the word "workflow" in a prompt to get started.
imagine thinking you invented an LLM talking to another LLM and it was in December 2025
In case you're curious about why dynamic workflows are so powerful and the future, read the RLM paper! Opus 4.8 + dynamic workflows in Claude Code is perhaps the first instance of a frontier model seriously trained to be an RLM.
I suspect within a year they'll just become the standard for nearly all coding agent interactions.
Claude Code is finally an RLM (oct 2025), congrats to Anthropic :-)
New in Claude Code (research preview): dynamic workflows.
Claude writes an orchestration script on the fly, then spins up a large fleet of coordinated subagents in parallel to take on your most complex tasks.
Use the word "workflow" in a prompt to get started.
Recently, I used dynamic workflows to catalogue all of our 100s of A/B test flags and find the ones rolled out to 0% or 100% so that we can quickly deprecate the stale ones. Instead of waiting for Claude Code to investigate each sequentially, dynamic workflows allowed Claude to process all of them in parallel in <10 minutes.
Excited to share our most powerful new Claude Code feature: dynamic workflows!
Mention "workflow" in a prompt and Claude will dynamically create an orchestration plan that it strictly follows, allowing you to confidently trust that every stage happens in the right order even across 100s of agents.
Claude dynamic workflows seemed pretty cool and I wanted an open-source version, so I made one. This one improves the test coverage of your entire repo, but you should be able to swap out the prompt and do anything else: https://github.com/OpenHands/software-agent-sdk/pull/3426
New in Claude Code (research preview): dynamic workflows.
Claude writes an orchestration script on the fly, then spins up a large fleet of coordinated subagents in parallel to take on your most complex tasks.
Use the word "workflow" in a prompt to get started.
sadly it's may 2026 and we still go through this weird cycle of making up claims & clearly not reading the paper. so let's clear this all up now (for the Nth time). I've made it explicitly clear time and time again that I don't claim I "invented" anything other than the actual RLM definition itself:
1. the paper doesn't claim it's the first to explore an LM talking to an LM (talking to an LM...) 2. the paper doesn't claim it's the first to explore sub-agents 3. the paper doesn't claim it's the first to argue for CodeAct-style execution 4. the paper doesn't claim it's the first propose recursion with respect to an LM 5. the paper doesn't claim to be the first to propose any of the independent features that make an RLM (context offloading, PTC, etc.)
all of the ideas that make up an RLM are somewhat intuitive and implemented in various ways, many with little success. the composition of them, and the lack of a need for any more, is what makes it unique. and the idea that it should be applied to arbitrary, systems independent of their use-case
the paper is an argument for exactly the type of abstractions one should define to better enable a sub-agent calling system (i.e. context offloading, CodeAct-style execution, programmatically sub-agent calling). It also argues that the line between a sub-agent calling system and a "language model" is blurry, and that a well-designed, general enough abstraction yields a meaningful "language model".
the last few releases in Claude Code have moved away from standard ReAct-style loops with JSON tool-calling to more of this abstraction. The recent release is very exciting to me (and why I talk about it being RLM-like) because a lot of the abstractions we've been arguing for are emerging in frontier lab systems.
tbh I get a lot of randos with takes like this, but as someone who quite likes the research you put out, this is pretty disappointing
imagine thinking you invented an LLM talking to another LLM and it was in December 2025
You can also set "/effort ultracode", a new effort level that runs at xhigh and lets Claude decide on its own when a task warrants a dynamic workflow.
More on how the orchestration works, and what early users have built with it, on the blog:
https://claude.com/blog/introducing-dynamic-workflows-in-claude-code
The Claude swarm will see you now
Excited to share our most powerful new Claude Code feature: dynamic workflows!
Mention "workflow" in a prompt and Claude will dynamically create an orchestration plan that it strictly follows, allowing you to confidently trust that every stage happens in the right order even across 100s of agents.
@lateinteraction wow man yeah they totally got that from you. they would've never thought of calling models in a loop. no chance
This is an idea I have been using for like 4 months now.
Very easy to do with -p or Agent SDK.
I doubt I will use CC for it, but great to see a native implementation of dynamic workflows.
Agent-to-agent interactions are super effective, but also watch out for token use.
New in Claude Code (research preview): dynamic workflows.
Claude writes an orchestration script on the fly, then spins up a large fleet of coordinated subagents in parallel to take on your most complex tasks.
Use the word "workflow" in a prompt to get started.
Anthropic valuation before deploying RLMs: $380B
Anthropic valuation after deploying RLMs: $965B
Use RLMs and you too can add $500B to your valuation
New in Claude Code (research preview): dynamic workflows.
Claude writes an orchestration script on the fly, then spins up a large fleet of coordinated subagents in parallel to take on your most complex tasks.
Use the word "workflow" in a prompt to get started.
We're like a year away from Claude instantly creating a C Corp and staffing it with 1000 virtual headcount overnight to solve any problem you put into the prompt box
Excited to share our most powerful new Claude Code feature: dynamic workflows!
Mention "workflow" in a prompt and Claude will dynamically create an orchestration plan that it strictly follows, allowing you to confidently trust that every stage happens in the right order even across 100s of agents.
to their credit, teams at anthropic have been unusually forthcoming repeatedly acknowledging the influence of RLMs and other open research on their thinking and design choices, e.g.
Claude Code is finally an RLM (oct 2025), congrats to Anthropic :-)
