TL;DR
Anthropic’s Claude Code team has introduced dynamic workflows, a feature that lets Claude write a JavaScript orchestration harness and coordinate temporary subagents during a single complex task. The company says the approach can improve parallel work, review, and judgment-heavy tasks, but it uses more tokens and is not meant for routine requests.
Anthropic’s Claude Code can now create dynamic workflows that assemble temporary teams of subagents for complex tasks, according to a June 2 company blog post cited by Thorsten Meyer AI. The development matters because it moves Claude Code beyond a single-agent work pattern toward task-specific orchestration, where separate agents can divide work, check one another, and merge results.
The feature, described by Anthropic as dynamic workflows, lets Claude write and run a small JavaScript harness for the task at hand. That harness can spawn subagents, coordinate their outputs, wait for parallel work to finish, and combine structured results into a final answer.
According to the Thorsten Meyer AI report, each subagent can receive a focused brief, a separate context window, and in some cases a different model choice. The stated goal is to reduce common single-agent failure modes: partial completion, self-review bias, and goal drift across long tasks.
Anthropic’s caveat is central to the announcement: this approach uses meaningfully more tokens and is aimed at complex, high-value tasks, not ordinary edits or short requests. The report frames the feature as useful when a task is large, parallel, adversarial, or judgment-heavy.
When one agent isn’t enough: Claude now builds its own team on the fly
Skills package what you know; loops decide how far you delegate over time. Dynamic workflows are the third axis — within a single task, Claude writes its own harness and assembles a temporary team of subagents. Think of it as Claude drawing an org chart for one job.
The shift is from prompting a worker to commissioning a team — more output, more cost, and a manager’s judgment required. Reach for a workflow when a task is big, parallel, adversarial, or judgment-heavy — and when you can feel a single agent getting lazy, grading its own homework, or losing the plot. Bound it (token budgets, pilot first) — workflows can spawn hundreds of agents and burn far more tokens. For everything else, don’t hire five people to change a lightbulb.
A Team Model for Claude Code
The change matters because many advanced agent tasks do not fail only from weak reasoning; they fail from work management problems. A single agent may stop before finishing a long checklist, favor its own prior answer, or lose a constraint after many steps. Dynamic workflows are meant to address those risks by separating execution, review, and synthesis.
For developers and technical teams, the feature could affect how they use Claude Code for large refactors, migrations, security reviews, research briefs, issue triage, and post-incident analysis. Anthropic’s reported patterns include fan-out-and-synthesize, adversarial verification, tournament-style judging, and loop-until-done workflows.
The trade-off is cost and control. A workflow that spawns many agents can consume far more tokens than a normal prompt. That makes budget limits, pilot runs, stop conditions, and careful task selection part of the practical use case, not an afterthought.
Workflow Automation with Microsoft Power Automate: Design and scale AI-powered cloud and desktop workflows using low-code automation
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
From Skills to Workflows
The Thorsten Meyer AI piece presents dynamic workflows as the third part of a broader Claude Code pattern. In that framing, skills package an organization’s knowledge, loops decide how work continues over time, and dynamic workflows decide how one complex task is divided among temporary agents.
The source report also emphasizes that this is not only a coding feature. Suggested applications include deep research, fact-checking claims, ranking large ticket backlogs, root-cause analysis, design review by rubric, model routing, and security tasks where separation of duties matters.
One security pattern highlighted in the report is quarantine: agents that read untrusted public content should be barred from high-privilege actions, while a separate agent handles action-taking. That structure is meant to reduce the risk that untrusted input can influence privileged operations.
“The feature is called dynamic workflows, and the plain description is that Claude writes its own harness.”
— Thorsten Meyer AI report
BUN PROJECTS: BUILD 7 REAL-WORLD JAVASCRIPT APPLICATIONS: Learn REST APIs, WebSockets, Databases & Deployment Through Hands-On Coding
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Costs and Limits Remain Open
It is not yet clear how widely developers will use dynamic workflows in production, how often the extra token cost will be justified, or how predictable these generated harnesses will be across varied tasks. The source material describes the feature as recent and still developing.
The report also does not provide independent benchmark results showing how much dynamic workflows improve outcomes over a single-agent approach. Claims about reduced bias, better completion, and stronger review are attributed to Anthropic’s described mechanics and the Thorsten Meyer AI analysis, not presented as independently verified performance results.
Building Agentic AI Systems with Claude Code: A Practical Guide to MCP, Sub-Agents, Hooks, Computer Use, and Production-Ready AI Workflows in 2026
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Developers Test the Workflow Boundary
The next step is practical adoption: teams using Claude Code will need to decide when a task merits multiple subagents and when a simpler prompt is enough. Early use is likely to focus on high-cost mistakes, such as security review, large code changes, and research tasks where verification matters.
Anthropic’s documentation at code.claude.com/docs is the main place to watch for implementation guidance, limits, and examples. For now, the safest reading is that dynamic workflows add a more powerful orchestration option, while leaving users responsible for scope, budget, and review.
AI Task Orchestration: Coordinating Complex Agent Workflows. A Comprehensive Guide to Building, Deploying, and Operating Multi-Agent AI Systems
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did Anthropic announce for Claude Code?
Anthropic says Claude Code can use dynamic workflows, where Claude writes a task-specific JavaScript harness that coordinates temporary subagents for complex work.
How is this different from using one Claude agent?
A single agent works inside one main context. Dynamic workflows can divide a task among separate subagents, each with a focused brief, then combine and review their outputs.
What kinds of tasks is this meant for?
The source material points to large refactors, migrations, deep research, fact-checking, security review, backlog triage, root-cause analysis, and other tasks that are large, parallel, or review-heavy.
What is the main drawback?
The main drawback is token use. Anthropic’s caveat, as reported, is that dynamic workflows consume meaningfully more tokens and are not meant for simple tasks.
What remains unproven?
Independent performance gains are still unclear. The reported benefits are tied to Anthropic’s mechanics and the Thorsten Meyer AI analysis, while broader production results are still developing.
Source: Thorsten Meyer AI
