Anthropic just changed what a single Claude Code session can accomplish. Dynamic workflows let Claude write its own orchestration scripts, spin up tens to hundreds of parallel subagents, and check its work before anything surfaces to you. The headline claim: work you would normally plan in quarters now finishes in days.
The core problem this solves is real. Some tasks are simply too big for a single agent in one pass. A bug hunt across an entire service, a migration touching hundreds of files, a plan you want stress-tested from every angle before you commit. A single-pass agent hits a ceiling fast, especially in complex or legacy codebases. Dynamic workflows are designed to handle exactly these end-to-end scenarios.
The key shift is that Claude generates the orchestration itself. You do not pre-define the workflow topology. Claude figures out what needs to run in parallel, writes the scripts, and coordinates the subagents. That is a meaningfully different model from static pipeline tools where you wire up the graph ahead of time.
Availability is already broad. Dynamic workflows are in research preview today across the Claude Code CLI, the Claude Code Desktop app, and the VS Code extension. They are also accessible via the Claude API, Amazon Bedrock, Vertex AI, and Microsoft (the source text is truncated there). Plan access covers Max, Team, and Enterprise tiers, with Enterprise requiring admin enablement.
What to do with this today: If you have a migration, large-scale refactor, or multi-file audit sitting in your backlog because it felt too big to hand off to an agent, this is the moment to run a test. Start with something bounded but genuinely multi-file, watch how Claude decomposes and parallelizes it, and see where the review checkpoints land. The research preview label means rough edges are expected, so pair it with a clean branch and treat the output as a strong first draft rather than a final commit.