# What Are Claude Dynamic Workflows? Up to 1,000 Agents, Explained

URL: https://whatisagenticai.net/blog/what-are-claude-dynamic-workflows/
Published: 2026-10-19
Publisher: Little Tough LLC, maker of AI Agent Assistant: Agentic AI (https://apps.apple.com/us/app/ai-agent-assistant-agentic-ai/id6761501260)

Claude dynamic workflows let Claude write a small program that runs many AI agents at once. What they are, where to use them, the limits, and what they cost.

Claude dynamic workflows are a way for Anthropic’s Claude to split a big job across many AI agents at once. Instead of one assistant grinding through a task step by step, Claude writes a small program that starts a crowd of helper agents, gives each a piece, and then checks and combines what they bring back.

The feature first showed up in Claude Code in May 2026. On October 9, Anthropic brought it to Claude Managed Agents, its platform for running agents in the cloud, and that is when the name started spreading. Here is what it means, in plain words.

**TL;DR:** Claude dynamic workflows are an Anthropic feature where Claude writes its own orchestration program to run many agents in parallel, then merges and verifies their results. They launched in Claude Code on May 28, 2026 (now generally available) and arrived in Claude Managed Agents as a beta on October 9, 2026, where one run can start up to 1,000 agents with up to 64 working at the same time. There is no extra fee for a run, but every agent uses tokens, so big runs can cost a lot.

## What are Claude dynamic workflows?

Think of a lead agent as a project manager. With a normal setup, the manager hands out tasks one by one, waits for each report, reads it, and then decides what to do next. That works, but the manager becomes the bottleneck.

With a dynamic workflow, the manager writes the plan down as code. Anthropic’s documentation describes a workflow as a program that runs many agents in phases and combines their results. The server then runs that program in the background as one "workflow run." The agents can work side by side, and the program handles who gets what and how the answers come back together.

The word "dynamic" matters. Nobody writes the workflow ahead of time. Claude writes a fresh one for each job, shaped around that specific task.

## When did dynamic workflows come out?

| Date | What happened |
| --- | --- |
| May 28, 2026 | Anthropic introduces dynamic workflows in Claude Code. The post has since been updated to say they are generally available. |
| June 2, 2026 | Anthropic publishes a follow-up guide, "A harness for every task," on patterns that work well with workflows. |
| October 9, 2026 | Dynamic workflows arrive in Claude Managed Agents as a beta, with runs of up to 1,000 agents. |

## How do Claude dynamic workflows work?

In Claude Code, you can start one in two ways, according to Anthropic’s launch post. You can simply ask Claude to "create a workflow." Or you can turn on a setting called ultracode from the effort menu, which raises the effort level and lets Claude decide on its own when a job is big enough to need a workflow. The first time, Claude Code shows what will run and asks you to confirm.

In Claude Managed Agents, developers switch it on in the agent’s settings (the multiagent_20261001 type, where workflows are on by default). The agent then decides when to start a run. Anthropic’s docs say its permission rules cover the tools the agents call, not the act of starting a run, so the main way to steer that choice is the agent’s instructions.

## Claude dynamic workflows limits and numbers

| Detail | Claude Managed Agents (beta) |
| --- | --- |
| Agents started per run | Up to 1,000 over the run’s life |
| Agents working at once | Up to 64 (Anthropic says this may change) |
| Default run length | 24 hours, unless the agent sets another lifetime |
| Open runs per session | 10 by default |
| Agents you can pre-list for a workflow | Up to 20 |
| Price of a run itself | None; you pay for every agent’s tokens |

For comparison, the older subagent setup in Managed Agents is one level deep and allows at most 25 child threads at a time. In Claude Code, Anthropic describes workflows as running tens to hundreds of parallel subagents in one session.

## Do dynamic workflows actually work better?

Anthropic gave two examples. In a bug hunt reported by The Decoder, Anthropic’s team hid 70 bugs in a codebase of about 116,000 lines. A single agent found between 14 and 27 per run. A dynamic workflow found 66 each time. That is Anthropic’s own test, and results on other kinds of work may differ.

The second example is bigger. In its launch post, Anthropic says Bun creator Jarred Sumner used workflows to port Bun from the Zig language to Rust: roughly 750,000 lines of Rust, 99.8% of the existing tests passing, and eleven days from first commit to merge. Anthropic notes that port was not yet in production.

Anthropic’s suggested uses are jobs with many pieces: bug hunts and security audits across a whole codebase, large migrations, deep research, and high-stakes work where independent agents try to break each other’s answers.

## Are Claude dynamic workflows free?

Not really. In Claude Code they come with Pro, Max, Team and Enterprise plans. Anthropic says they are on by default for Max, Team and Enterprise, and Pro users can switch them on in settings. They also work through the Claude API, Amazon Bedrock, Vertex AI and Microsoft Foundry.

The catch is usage. Anthropic warns that workflows can consume "substantially more tokens than a typical Claude Code session" and suggests starting with a small, scoped task. In Managed Agents you can set a session budget. When it runs out, open runs pause.

## Dynamic workflows vs subagents vs agent teams

Subagents are a manager handing out tasks and reading each report. Dynamic workflows are a manager writing a script that runs the whole crew at once. Agent teams, the kind you find in consumer apps, are a small group of specialists who read each other’s work in one conversation, which we explain in [why two AI agents beat one](https://whatisagenticai.net/blog/multi-agent-teams/). All three rest on the same basic [agent loop](https://whatisagenticai.net/blog/how-ai-agents-work/). They differ in scale and in who coordinates.

## What this means for you

Dynamic workflows are a developer tool. They live in a coding app and a cloud platform, and they can burn through a lot of tokens. Most people will never start a 1,000-agent run.

The idea underneath them is the useful part: several focused agents, checking each other, beat one agent doing everything alone. You do not need a terminal to get that. **Agentic AI** puts a team of ready-made expert agents on your iPhone and iPad. You pick up to four, such as a Research Analyst, a Writing Coach and a Life Planner, give them one task, and each builds on what the others wrote. There is no account and no login, and the first 3 tasks are free.

## Quick answers

**What is a dynamic workflow in Claude?** It is a program Claude writes for one job that runs many agents in the background, often in parallel, and then combines and checks their results.

**How do Claude dynamic workflows work?** A lead agent plans the job, writes an orchestration program, and the server runs it as a workflow run. Each helper agent handles one piece, and the program merges the answers.

**What are Claude Code dynamic workflows?** The same feature inside Claude Code, launched May 28, 2026. You ask Claude to create a workflow or turn on ultracode, and Claude asks you to confirm the first time.

**How many agents can a Claude dynamic workflow run?** In Claude Managed Agents, up to 1,000 agents per run, with up to 64 working at once. In Claude Code, Anthropic says tens to hundreds of parallel subagents.

**Claude dynamic workflows vs agent teams?** Workflows run many agents from a program, mostly for big coding and research jobs. Agent teams are a few specialists working in one shared conversation, which suits everyday tasks better.

New to these words? Our [agentic AI glossary](https://whatisagenticai.net/agentic-ai-glossary/) covers them in plain English.

**Sources:** [Anthropic: Introducing dynamic workflows in Claude Code](https://claude.com/blog/introducing-dynamic-workflows-in-claude-code) · [Claude Platform Docs: Multiagent orchestration](https://platform.claude.com/docs/en/managed-agents/multiagent-orchestration) · [The Decoder](https://the-decoder.com/anthropics-claude-can-now-orchestrate-up-to-1000-ai-agents-in-parallel-through-dynamic-workflows/) · [MIXED](https://mixed-news.com/en/claude-managed-agents-dynamic-workflows-1000-agents-permission-policies/)