Claude Code adds dynamic workflows (research preview) for large parallel agent runs

Anthropic shipped dynamic workflows in Claude Code as a research preview, designed for unusually large engineering tasks like codebase-wide audits and migrations.

What’s new

Dynamic workflows let Claude write an orchestration script on the fly, break work into subtasks, and then run tens to hundreds of parallel subagents in a single session, verifying results before reporting back (Claude blog).

Anthropic positions this for problems that are difficult to complete with a single agent pass, including:

• Codebase-wide bug hunts, optimization audits, and security audits (Claude blog)
• Large migrations and modernization efforts across thousands of files (Claude blog)
• High-stakes work where you want independent attempts and adversarial checking before you see results (Claude blog)

Availability

Dynamic workflows are available today in research preview in the Claude Code CLI, Desktop, and VS Code extension for Max, Team, and Enterprise (if admin enabled) plans, and also via the Claude API plus major cloud platforms including Amazon Bedrock, Vertex AI, and Microsoft Foundry (Claude blog).

How to try it

Anthropic says you can start a workflow in two main ways:

1. Ask Claude to create one directly (for example, “Create a workflow”), or
2. Enable the Claude Code setting ultracode, which sets effort to xhigh and lets Claude decide when to use a workflow automatically (Claude blog).

Notes on usage

Anthropic notes that dynamic workflows can consume substantially more tokens than a typical Claude Code session, and the first time a workflow triggers Claude Code will show what’s about to run and ask you to confirm (Claude blog).