What Are Dynamic Workflows?

Dynamic workflows are a new capability in Claude Code that enable the AI to tackle very large-scale problems by intelligently decomposing them into smaller tasks and executing those tasks in parallel. Think of it as Claude becoming a technical lead that plans the work, assigns tasks to a team of subagents, monitors progress, and reviews the final output.

This is a significant departure from the standard conversational model where each response is generated sequentially. With dynamic workflows, Claude can:

• Analyze a large problem and create a structured plan
• Spawn hundreds of parallel subagents, each working on a distinct piece of the puzzle
• Monitor subagent progress and collect results
• Verify outputs against quality criteria (e.g., an existing test suite)
• Synthesize results and report back with a coherent summary

How Dynamic Workflows Work

The dynamic workflows architecture operates in several phases:

1. Planning Phase

Claude analyzes the user’s request and creates a work plan. For a codebase migration, this means understanding the current state, the desired target state, and breaking the migration into independent chunks that can be processed in parallel.

2. Orchestration Phase

Claude spawns subagents — each one an independent Claude Code session — and assigns them specific tasks. The orchestration layer manages communication, resource allocation, and progress tracking across all subagents.

3. Execution Phase

Subagents execute their assigned tasks in parallel. Because they operate independently, there is no sequential bottleneck. With hundreds of subagents running simultaneously, the total throughput is dramatically higher than what a single conversational session could achieve.

4. Verification Phase

Before reporting back, Claude verifies subagent outputs. For code changes, this means running the existing test suite against changes to ensure nothing is broken. For other tasks, verification might involve consistency checks, format validation, or cross-referencing.

5. Synthesis Phase

Finally, Claude synthesizes all results into a coherent report for the user, highlighting what was changed, what was verified, and any issues that need human attention.

Hundreds of Parallel Subagents

The headline capability of dynamic workflows is the ability to run hundreds of parallel subagents in a single session. This is not merely a theoretical limit — early testers are already using this to accomplish tasks that would be impractical with traditional AI coding tools.

Each subagent operates as an independent Claude Code session with its own context, file access, and task assignment. The orchestration layer ensures that subagents don’t conflict with each other and that their outputs can be merged coherently.

This approach is particularly powerful for:

• Large-scale refactoring — Each subagent handles a module or file
• Test generation — Hundreds of subagents generate tests for different components simultaneously
• Documentation updates — Each subagent updates documentation for its assigned area
• Dependency upgrades — Subagents handle different dependency trees in parallel

Codebase-Scale Migrations

The marquee use case for dynamic workflows is codebase-scale migrations. Anthropic specifically highlighted that Claude Code with Opus 4.8 and dynamic workflows can now carry out migrations across hundreds of thousands of lines of code — from kickoff to merge, with the existing test suite as its bar.

Consider a migration from one framework to another (e.g., React Class components to hooks, or JavaScript to TypeScript). With traditional AI coding tools, this would require either:

• Countless iterative prompts, each handling one file at a time
• A custom script that lacks the intelligence to handle edge cases

Dynamic workflows change this entirely. Claude plans the migration, spawns subagents for each file or module, executes all changes in parallel, runs the test suite to verify nothing broke, and presents the completed migration for review. What once took days can now happen in minutes.

How Opus 4.8 Enhances Dynamic Workflows

Dynamic workflows are available in Claude Code regardless of which model you use, but they are particularly powerful with Opus 4.8 for two reasons:

• Longer agent runtimes: Opus 4.8 subagents can run for extended durations without degrading in quality, making them suitable for complex multi-step tasks within the workflow
• Better judgment: Opus 4.8’s improved honesty and self-correction mean subagents are more likely to flag issues in their own work rather than silently producing flawed output

Real-World Use Cases

Beyond migrations, dynamic workflows enable a range of large-scale engineering tasks:

• Dependency upgrades: Upgrade a library across hundreds of files, handling API changes and deprecation warnings in parallel
• Test suite expansion: Generate comprehensive unit tests for an entire codebase in a fraction of the time
• Code standardization: Enforce consistent patterns, formatting, and best practices across thousands of files
• Documentation generation: Create and update documentation for every module in parallel
• Bug fixing: Identify and fix similar bug patterns across the entire codebase simultaneously
• Performance optimization: Profile and optimize different modules in parallel

Availability & Plans

Dynamic workflows are currently available as a research preview in Claude Code for Enterprise, Team, and Max plans. This means the feature is functional and being used by early adopters, but Anthropic may continue refining it based on feedback.

To access dynamic workflows, users need:

• A Claude Code subscription on Enterprise, Team, or Max
• Claude Code installed and configured
• Access to the research preview (rolling out to eligible accounts)

Anthropic has indicated that dynamic workflows will continue to evolve, with improvements to orchestration, subagent coordination, and verification capabilities expected in future updates.

Frequently Asked Questions