Anthropic's Claude Code Deploys Hundreds of Parallel Agents to Rewrite Entire Codebases
Dynamic workflows in Claude Code go GA, while scheduled deployments and credential vaults hit public beta in Claude Managed Agents
Takeaways
- Dynamic workflows GA: Claude Code can now self-orchestrate tens to hundreds of parallel subagents for complex, multi-hour engineering tasks.
- Adversarial verification built in: Subagents independently attempt problems and try to refute each other's findings before results reach the user.
- Bun rewrite proof point: Jarred Sumner ported ~750,000 lines from Zig to Rust in 11 days with 99.8% test suite passing using dynamic workflows.
- Scheduled deployments (public beta): Claude Managed Agents can now run on a cron schedule with no scheduler infrastructure to build or host.
- Credential vaults (public beta): API keys are injected at the network boundary so agents can use authenticated CLIs without the model ever seeing the secret.
- Token cost warning: Dynamic workflows consume significantly more tokens than standard sessions; Anthropic recommends starting with scoped tasks.
Anthropic used its Code with Claude Tokyo event to push three significant updates to its agentic platform at once: dynamic workflows in Claude Code are now generally available, and two new Claude Managed Agents features -- scheduled deployments and environment variables in vaults -- have entered public beta. Together, they represent a meaningful shift in what Claude can do without a human in the loop.
Fleets of agents, writing their own playbook
Dynamic workflows let Claude tackle the most challenging tasks end-to-end by dynamically writing orchestration scripts that run tens to hundreds of parallel subagents in a single session, checking its work before anything reaches you. The key word here is dynamic: Claude isn't following a fixed pipeline you designed. It plans the breakdown of work itself, based on your prompt.
Some problems are too big for one pass by a single agent -- a bug hunt across an entire service, a migration that touches hundreds of files, a plan you want stress-tested from every angle before you commit to it. Dynamic workflows are designed to absorb all of that.

When a workflow kicks off, Claude plans dynamically based on your prompt, breaks it into subtasks, and fans the work out across subagents running in parallel. Results are checked before they're folded in, and you come back to a single, coordinated answer. Agents can also work adversarially -- other agents try to refute what they found, and the run keeps iterating until the answers converge. This is how the system catches errors that a single-pass agent would miss.
Dynamic workflows are built for parallel and long-running work that can extend into hours and days. Progress is saved as the run goes, so a job that's interrupted picks up where it left off instead of starting over.
The Bun rewrite: a proof of scale
The most striking real-world demonstration of what dynamic workflows can do is the recent Bun rewrite. Jarred Sumner used dynamic workflows to port Bun from Zig to Rust with 99.8% of the existing test suite passing, roughly 750,000 lines of Rust, and eleven days from first commit to merge.
One workflow mapped the right Rust lifetime for every struct field in the Zig codebase. The next wrote every .rs file as a behavior-identical port of its .zig counterpart, hundreds of agents working in parallel with two reviewers on each file. A fix loop then drove the build and test suite until both ran clean. That's not a demo -- it's 750k lines of production-grade code, verified end-to-end.
What it's actually good for
Early access users have converged on a few high-value patterns:
- Codebase-wide audits: Claude searches a service or repo in parallel, then runs independent verification on every finding so the report surfaces real issues -- auth checks, input validation, and unsafe patterns across an entire codebase.
- Large migrations: Claude can handle framework swaps, API deprecations, and language ports that span thousands of files end-to-end.
- High-stakes verification: When the cost of a wrong answer is high, a workflow gives Claude independent attempts at the problem and adversarial agents working to break the result before you see it.
Klarna's engineering team noted that dynamic workflows were "especially valuable for discovery and review tasks across large codebases,