Anthropic announced Dynamic Workflows for Claude Code, a feature that shifts coordination out of conversational context and into executable JavaScript orchestration. Claude generates a bespoke JS script for the specific task, distributes work across hundreds of parallel subagents, and iteratively verifies results until they converge.
Use cases
Dynamic Workflows targets problems too large or complex for a single agent pass: for example, bug hunts across entire services, migrations that touch hundreds of files, or plans you want stress-tested from multiple angles.
Key elements of the approach include:
- Claude as the orchestrator: there are no prebuilt templates; Claude writes and runs a custom JavaScript orchestrator tailored to the task.
- Independent verification: agents attack the problem from different angles, other agents attempt to refute findings, and runs iterate until results align.
- Resumable and saveable: progress is checkpointed so interrupted jobs can resume, and workflows can be saved as reusable /command entries with structured input parameters.
Proof of concept: porting Bun
Jarred Sumner used Dynamic Workflows to port the Bun project from Zig to Rust. The result was 750,000 lines of Rust, with 99.8% of the test suite passing, and the timeline from first commit to merge was eleven days. Hundreds of agents worked in parallel, with two human reviewers assigned per file.
Warnings and availability
- Token usage: these workflows consume meaningfully more tokens than a typical session. It is recommended to start with a scoped task to understand usage before applying the workflow to an entire codebase.
- Availability: the Dynamic Workflows research preview is available on Max, Team, and Enterprise plans, as well as via the Claude API, Amazon Bedrock, Vertex AI, and Microsoft Foundry.
- Requirements: Claude Code v2.1.154+ is required. You can enable the ultracode effort level or ask Claude to "create a workflow" to start.
By encoding coordination into executable scripts and building independent verification into runs, Dynamic Workflows aims to make large-scale, multi-step software tasks repeatable, resumable, and parallelizable.
