Dynamic Workflows - Dynamic Workflows Launched by Claude Code

What Is Dynamic Workflows

Dynamic Workflows is Claude Code 's dynamic workflow, currently in research preview, supporting Claude the dynamic authoring of orchestration scripts to schedule dozens to hundreds of sub-agents in parallel within a single session, automatically decompose tasks, verify results, and resume from checkpoints, compressing large engineering tasks that would originally take weeks into completion within days.

Key Features of Dynamic Workflows

Automatic Task Decomposition and Parallel Execution: Claude automatically breaks down large tasks into multiple sub-tasks based on user instructions and dispatches them to parallel sub-agents for simultaneous processing.
Result Verification and Adversarial Review: The results of each sub-agent undergo independent verification before being aggregated, while adversarial agents are deployed to actively challenge existing conclusions, ensuring convergence of answers.
Progress Checkpoint and Resume: Progress is automatically saved during workflow execution, allowing recovery from checkpoints after interruptions without starting over.
Full-Process Visualization: Real-time display of the model used by each sub-agent, token consumption, number of tool calls, and execution time.

Technical Principles of Dynamic Workflows

Dynamic Orchestration Script Generation: Claude generates coordination scripts in real time based on task characteristics rather than relying on predefined templates, enabling flexible task decomposition and scheduling.
Multi-Level Verification Mechanism: Adopts a generate-review-iterate loop, where multiple agents independently attempt problems from different angles while other agents are responsible for rebuttal and verification, until results converge.
External Coordination Architecture: Coordination logic runs outside the conversation, ensuring that plans can proceed steadily even as task scale expands, without being limited by conversational context.
Long-Running Support: Designed to support continuous operation for hours to days, adapting to the time requirements of large engineering tasks.

How to Use Dynamic Workflows

Start with Direct Instructions: In Claude Code , directly ask Claude to create a workflow; for example, entering "Create a workflow" triggers automatic orchestration.
Enable ultracode Mode: Enable via the Effort menu ultracode setting, which automatically sets effort to xhigh, and Claude will autonomously determine when to trigger workflows to handle complex tasks.
Platforms and Permissions：
- Applicable to Claude Code CLI, desktop, VS Code extension, as well as Claude API, Amazon Bedrock, Vertex AI, and Microsoft Foundry.
- Enabled by default for Max, Team, and API users; Enterprise users require an administrator to manually enable it in settings.

Core Advantages of Dynamic Workflows

Scalable Processing Capability: A single session can coordinate hundreds of agents working in parallel, suitable for large-scale codebase-level operations across files and services.
High-Reliability Output: Multiple verification and adversarial review mechanisms significantly reduce error rates, suitable for critical tasks in scenarios where mistakes are costly.
End-to-End Automation: The entire process from task planning to implementation verification is completed automatically, without manual intervention in orchestration or sub-agent management.
Real-World Verification Cases: Jarred Sumner used Dynamic Workflows to port Bun from Zig to Rust, generating about 750,000 lines of code with a test suite passing at 99.8%, taking only 11 days from first commit to merge.

Dynamic Workflows project address

Project official website：https://claude.com/blog/introducing-dynamic-workflows-in-claude-code

Comparison of Dynamic Workflows with similar competitors

Dimension	Claude Code Dynamic Workflows	OpenAI Codex CLI
Parallel scale	Hundreds of sub-agents run in large-scale parallel, suitable for codebase-level refactoring	Sub-agent parallel scale is limited, more suitable for module-level tasks
Task orchestration	Dynamically and automatically generates orchestration scripts without manual intervention	Requires manual triggering or configuration by the user, with a lower degree of orchestration automation
Verification mechanism	Built-in adversarial review and multi-round iterative verification, high reliability	Relies on Auto-review and user approval, with a more traditional verification process
Runtime duration	Natively supports multi-day long-running execution with resumption from checkpoints	Codex Cloud supports long-running background execution, but local sessions are limited
Token efficiency	Higher consumption (large multi-agent parallel overhead)	Low token consumption (about 1/3 to 1/4 of Claude)
Terminal performance	Leads on complex multi-file tasks such as SWE-bench Pro	Leads on terminal programming tasks such as Terminal-Bench 2.0
Usage mode	Emphasizes "cyber contractor" style collaboration, with developers able to see the full process	Emphasizes "fire-and-forget" unattended operation, automatically completed in the background

Application scenarios of Dynamic Workflows

Full codebase vulnerability scanning: Searches the entire service or repository in parallel, runs independent verification on each finding, and outputs a report of real issues.
Large-scale code migration: Handles end-to-end migrations involving thousands of files, such as framework replacement, API deprecation, and language porting.
Dual verification for critical tasks: Before high-risk decisions, ensures reliable conclusions through independent attempts and adversarial challenges.
Performance optimization audit: Based on profiler guidance, reviews dead code, performance bottlenecks, and cleanup opportunities in the codebase in parallel.