The feature automates complex migrations using parallel subagents.
Also new in Claude Code: dynamic workflows (research preview).
For the hardest tasks, Claude makes a plan, runs hundreds of parallel subagents, and verifies its work before reporting back. Think a migration touching hundreds of files.
Read more: http://claude.com/blog/introducing-dynamic-workflows-in-claude-code
Users praise Claude Code's dynamic workflows with parallel subagents for automating complex tasks and providing productivity gains like editing systems in one pass, but some dismiss it because of prompt limits.
Dynamic workflows and adversarial code review was part of what made it possible to rewrite Bun in Rust in 6 days.
New in Claude Code (research preview): dynamic workflows.
Claude writes an orchestration script on the fly, then spins up a large fleet of coordinated subagents in parallel to take on your most complex tasks.
Use the word "workflow" in a prompt to get started.
Normal code review: Find bugs. Look harder and find more. 45 "bugs" later -- how do you know which are real?
Adversarial code review: Find bugs. Keep going. For each bug, ask 2+ Claudes in isolated context windows to refute - is this bug real? Only the real ones survive.
Adversarial reviews improve accuracy, but cost more tokens. Ask CC workflows to run an adversarial review and it'll do the right thing!
@jarredsumner Can you elaborate on adversarial code reviews? Is this a specialized cc agent? Is this documented? Couldn’t find anything
Example prompt:
Write a workflow where for each finding in ./REPORT.md.
Step 1: Fix the bug. Do not use any git or build commands to avoid stepping on another Claude running in the same branch. Step 2: Use 2 adversarial review agents to refute the bugfix. Uncover every flaw. Do not use any git or build commands to avoid stepping on another Claude running in the same branch.
After all are complete:
Step 3: Apply all the bugfixes, run the build and get the relevant tests to pass. Once passing, commit and make a PR.
Start by splitting up the ./REPORT.md into individual files with bugs and pass these files to the workflow.
A pattern that works well
For each unit of work: 1) Do the work. Don’t use git / cargo, slow commands are banned 2) Adversarial review 3) Apply changes
Examples: - Rewrite every .zig file to .rs following patterns in ./PORTING and ./LIFETIMES.tsv - Get the crates to compile
A pattern that works well
For each unit of work: 1) Do the work. Don’t use git / cargo, slow commands are banned 2) Adversarial review 3) Apply changes
Examples: - Rewrite every .zig file to .rs following patterns in ./PORTING and ./LIFETIMES.tsv - Get the crates to compile
Dynamic workflows, in my personal opinion, is the state of the art today for reliably using agents to complete medium - large projects.
For extra large projects, I especially like using workflows with /loop.
@claudeai If you don't understand what is meant by dynamic workflow, here I've made a video on why this matters to you.
Credits:- Created using Distilbook.
Claude Opus 4.8 is available today on the web, the Claude Platform, and all major cloud platforms.
Read more: http://anthropic.com/news/claude-opus-4-8
(still writing blog post, this isn’t that)
Dynamic workflows and adversarial code review was part of what made it possible to rewrite Bun in Rust in 6 days.
Claude Opus 4.8 is available today on web, the Claude Platform, and all major cloud platforms.
Read more: http://anthropic.com/news/claude-opus-4-8
Workflows can cost a lot of tokens. And that plus the split context windows are part of what makes them effective.
You can ask Claude to use a ballpark number of tokens to use more for harder tasks or to save tokens for easier tasks
When you split up the work this way, Claude focuses on just the task it needs to and empirically this makes it more effective.
I’ve seen individual workflows run for 10 hours continuously. I imagine this will be very common.
Each individual step within a workflow gets the same prompt with arguments, and then within those steps a different context window (or a fork). Workflows split up work more deterministically than subagents. It’s closer to a bespoke build system for a project than chat.
Example prompt:
Write a workflow where for each finding in ./REPORT.md.
Step 1: Fix the bug. Do not use any git or build commands to avoid stepping on another Claude running in the same branch. Step 2: Use 2 adversarial review agents to refute the bugfix. Uncover every flaw. Do not use any git or build commands to avoid stepping on another Claude running in the same branch.
After all are complete:
Step 3: Apply all the bugfixes, run the build and get the relevant tests to pass. Once passing, commit and make a PR.
Start by splitting up the ./REPORT.md into individual files with bugs and pass these files to the workflow.
When you split up the work this way, Claude focuses on just the task it needs to and empirically this makes it more effective.
I’ve seen individual workflows run for 10 hours continuously. I imagine this will be very common.
Each individual step within a workflow gets the same prompt with arguments, and then within those steps a different context window (or a fork). Workflows split up work more deterministically than subagents. It’s closer to a bespoke build system for a project than chat.
Dynamic workflows, in my personal opinion, is the state of the art today for reliably using agents to complete medium - large projects.
For extra large projects, I especially like using workflows with /loop.
@claudeai Really cool feature! I recorded a video about it last week :)
@claudeai Claude Opus 4.8 is here! Sharper judgment, longer independent runs, and dynamic workflows with parallel subagents? Game-changer for big coding projects. Fast mode at 2.5x speed & 1/3 cost is Anthropic leveling up again! Who's trying it first?
@claudeai Claude's dynamic workflows sound game-changing Hundreds of parallel subagents planning, executing & verifying massive refactors? This is the agentic leap we've been waiting for. Anthropic is pushing boundaries hard. Time to test on a real codebase! cc @grok @xai
@claudeai Linear context doesn't scale. Hierarchical context does.
Dynamic workflows are proving this for agents; while PageIndex is proving it for retrieval.
The future isn't infinite context. It's agentic, adaptive context trees.
https://github.com/VectifyAI/PageIndex
@0xRapid
@claudeai Anthropic always saying we don’t need the complex setups we have, and then every release adding parts of our complex setups.
I’ve had dynamic workflows in my ai agent for the last 9 months.
@claudeai Seeing the dynamic workflow of Claude Opus 4.8, now AI has actually become an engineer! Parallel agents + self-verification = future aa went brother. Congrats Anthropic team!
@jarredsumner I get this error ⎿ API Error: 400 messages.3.content.8: `thinking` or `redacted_thinking` blocks in the latest assistant message cannot be modified. These blocks must remain as they were in the original response. Whyy?
