Overview
Dynamic Workflows is a new feature of Claude Code launched by Anthropic on May 28, 2026, alongside Claude Opus 4.8, with a HN 198-point hot post. It upgrades Claude Code from 'a single AI working alone' to 'one AI orchestrating dozens to hundreds of AIs collaborating.'
The usage is simple: you input a sentence in Claude Code with the word 'workflow.' Claude no longer works alone but first writes a scheduling script, then dispatches dozens or even hundreds of sub-agents to work separately and review each other, finally merging the results for delivery. This is an official native Multi-Agent solution for large-scale engineering tasks, similar in positioning to open-source Multi-Agent frameworks like OpenHands / SWE-Agent but more deeply integrated.
Key Features
- One-sentence launch of multi-agent collaboration: No manual configuration needed; trigger by inputting a command with 'workflow,' and Claude plans the scheduling script itself
- Dozens to hundreds of sub-agents in parallel: Can run hundreds of sub-agents simultaneously, each responsible for a small task, with the main agent handling integration
- Mutual review among agents: Supports agent mutual evaluation + voting mechanism, reducing single-point output errors and improving stability for large-scale projects
- Deep integration with Opus 4.8: Relies on Opus 4.8's long context and stable output, allowing the main agent to handle large-scale codebase transformations stably
- Full observability: The Claude Code terminal displays the progress, status, and logs of each sub-agent, allowing manual intervention or termination
Use Cases
- Large-scale code migration: refactoring / framework upgrades across hundreds of files
- Full-repository test coverage improvement: dispatching dozens of agents to write tests in parallel
- Large PR review: each agent reviews a portion of code, then collectively deliberates
- Cross-service code synchronization refactoring: each agent responsible for one service
- Documentation generation: each agent writes documentation for one module, with the main agent merging and proofreading
Pros
- Makes multi-agent collaboration a native feature rather than an external plugin, with extremely low startup cost
- Mutual review mechanism significantly improves stability for large-scale tasks
- Leverages Opus 4.8, ensuring the main agent remains stable in long contexts without dropping the ball
- Full observability offers more control than black-box multi-agent frameworks
- For developers, it's an example of packaging 'Harness engineering' capabilities directly for delivery
Pricing
Requires a paid Claude API account or a Claude Pro / Max subscription. Multi-agent parallelism significantly increases token consumption; running hundreds of sub-agents in one round may consume millions to tens of millions of tokens—heavy usage is recommended via API billing with Prompt Caching optimization.
Summary
Dynamic Workflows is primarily aimed at developers and teams who 'really have large-scale engineering tasks'—scenarios like framework migration, cross-repository refactoring, and codebase-level restructuring. For daily small projects by a single person, it's unnecessary; a single agent suffices. Its true value lies in lowering multi-agent collaboration from 'advanced play' to a native capability usable with 'one command,' serving as Claude Code's direct response to competitors like Cursor / Devin in terms of Harness thickness.