Today Anthropic announced something that made me sit up — and at the same time confirmed what we've been doing at Amazing Agents for months.
It's called "Dynamic Workflows" in Claude Code. The principle: instead of a single AI agent working through a task alone, Claude writes an orchestration script at runtime and uses it to steer dozens to hundreds of parallel subagents within a single session. The plan becomes executable code, not just conversation.
What's new?
Claude now decides for itself when a dynamic workflow makes sense. It builds an executable plan, distributes work across specialized subagents, checks intermediate results, and assembles everything — before any result reaches you. Progress is saved: interrupted runs resume at the checkpoint instead of starting over. For critical tasks, the system can even launch independent solution attempts and cross-checks of each other's quality — built-in verification against errors.
Anthropic names concrete use cases:
- Codebase-wide security audits: scanning hundreds of files in parallel for vulnerabilities.
- Large migrations: framework switches, API deprecations, even language ports across thousands of files.
- Adversarial reviews: two agents check the same plan from opposing perspectives before a commit happens.
- Profiler-guided performance optimization: systematic analysis of entire codebases for bottlenecks.
The most impressive number: Jarred Sumner used Dynamic Workflows to port Bun from Zig to Rust — 750,000 lines of code, 99.8% of existing tests passing, eleven days from first commit to merge. What usually takes months became a sprint.
Why this interests me as an agency founder
This exact pattern — an orchestrator steering specialized agents in parallel — is what we've been building at Amazing Agents for clients for months. Not for code, but for business processes in mid-sized companies.
Translated into our clients' day-to-day, it looks like this:
- Security audit → document check: an agent checks hundreds of documents in parallel for completeness and compliance — instead of a case worker clicking through folders.
- Migration → multi-quote comparison: several specialized agents fetch quotes simultaneously, normalize the data, and deliver a decision template.
- Adversarial review → quality check before sending: before a quote goes out, a second agent checks for errors, inconsistencies, and missing information.
What Anthropic is now showing developers is our mid-sized-business routine: a digital employee coordinating several others.
Klarna leads the way
Alessio Vallero, Senior Engineering Manager at Klarna, on Dynamic Workflows:
"Dynamic workflows have been especially valuable for discovery and review tasks across large codebases. We've seen strong results using it to identify dead code and surface cleanup opportunities that traditional static analysis missed."
When Sweden's largest fintech relies on parallel agent workflows in production, this is no longer an experiment — it's standard enterprise tooling. Proof that the technology is production-ready. Mid-sized business doesn't need the same infrastructure. Mid-sized business needs the right agents.
What you should know
Dynamic Workflows is currently a research preview — available in the Claude Code CLI, Desktop, and the VS Code extension for Max, Team, and Enterprise plans. Anthropic itself points out that the workflows use "substantially more tokens" than a typical session. The technology is here. But it isn't trivial. It needs the right orchestration, the right specialists, and someone who brings both together.
That's exactly our job.
Want to know which of your processes would benefit from parallel agent teams? Let's talk.