Claude Code’s Dynamic Workflows Take on the Tasks That Were Too Big to Automate

Anthropic has introduced dynamic workflows in Claude Code, and the feature addresses a problem that most development teams know well: some engineering tasks are simply too large and complex for a single AI agent to handle in a single pass.

With dynamic workflows, Claude Code can now write its own orchestration scripts and spin up tens to hundreds of parallel subagents within a single session. Those subagents divide the work, check each other’s findings, and hand you a single coordinated result — without requiring you to build out a full agent pipeline yourself.

The feature is available today in research preview across the Claude Code CLI, Desktop and VS Code extension for Max, Team and Enterprise plans. It’s also accessible through the Claude API, Amazon Bedrock, Vertex AI and Microsoft Foundry.

What it’s Built For

Dynamic workflows are designed for tasks where scope is the limiting factor. Think about a bug hunt across an entire service, a migration that touches hundreds of files, or a security audit where you need every finding independently verified before it lands in a report. These are the kinds of tasks that teams often break into sprints, assign to multiple engineers, and plan across quarters.

According to Anthropic, that same class of work can now run in days.

A few categories stand out as early use cases:

• Codebase-wide audits: Claude searches a service or repository in parallel, then runs independent verification on every finding. The same pattern works for security hardening — checking authentication, input validation, and unsafe patterns across an entire codebase.
• Large migrations: Framework swaps, API deprecations, and language ports that span thousands of files can be handled end-to-end.
• High-stakes verification: When the cost of a wrong answer is high, workflows assign independent agents to attack the same problem from different angles. Adversarial agents actively try to break the result before it reaches you.

Early access users are already seeing results. Teams at Klarna used dynamic workflows to identify dead code and surface cleanup opportunities that traditional static analysis missed. Engineers at CyberAgent noted that the feature bridges the gap between firing off a single subagent and managing a full agent team — allowing plans to flow into implementation without losing visibility.

One of the more striking examples comes from Jarred Sumner, who used dynamic workflows to port Bun from Zig to Rust. The result: roughly 750,000 lines of Rust, with 99.8% of the existing test suite passing, completed in eleven days from first commit to merge. Separate workflows mapped Rust lifetimes for every struct field, wrote behavior-identical .rs ports of every .zig file in parallel, and then drove a fix loop until the build and test suite ran clean. An overnight workflow was implemented to eliminate unnecessary data copies and open a PR for each one for final review.

How it Works

When a workflow starts, Claude dynamically plans based on your prompt, breaks the task into subtasks, and distributes the work across parallel subagents. Results are checked before they’re folded in, and the run keeps iterating until findings converge — producing results that a single pass can’t achieve.

Progress is saved as the run goes. If a job is interrupted, it picks up where it left off rather than starting over. The coordination happens outside the conversation, so the plan stays on track regardless of how large the task gets.

It’s worth noting: dynamic workflows consume substantially more tokens than a typical Claude Code session. Anthropic recommends starting with a scoped task to get a feel for your usage patterns before scaling up.

Mitch Ashley, VP and practice lead for software lifecycle engineering at The Futurum Group, sees the verification challenge as the critical issue teams need to solve before scaling this up. “Parallel agent orchestration moves the hard problem from writing code to confirming it is correct,” he said. “When one session fans out into hundreds of subagents across a migration or an audit, change outpaces what a team can review by hand. Organizations adopting this need verification, governance, and evidence capture that scale at the pace of generation. Treat that as an afterthought and you inherit verification debt faster than the tooling pays it down.”

Getting Started

For Max and Team plan users, or anyone accessing Claude Code through the API, dynamic workflows are on by default. You have two options to start a workflow: Ask Claude directly to create one (for example, “Create a workflow”), or enable the Ultracode setting through the effort menu. The ultracode setting bumps effort to the highest level and lets Claude decide automatically when a workflow is the right tool for your task. Turning on auto mode before starting is recommended for the best experience.

Enterprise plan users will find dynamic workflows off by default at launch. Admins can enable them through the Claude Code managed settings.

Full documentation is available at code.claude.com.

Dynamic workflows represent a meaningful step in what AI coding tools can actually achieve end-to-end. For DevOps and platform engineering teams dealing with legacy modernization, large-scale migrations, or recurring audit work, the question is no longer whether AI can help — it’s how much of the coordination overhead you’re willing to hand off. But as Ashley notes, the teams that move fastest will be the ones that treat verification and governance as first-class problems, not follow-on tasks.

from DevOps.com https://ift.tt/ixDHIJE