TL;DR
Anthropic has described a Claude Code capability called dynamic workflows, in which Claude writes a task-specific JavaScript harness to coordinate subagents. The company presents it as a tool for complex, high-value work, while warning that it can use far more tokens than ordinary single-agent prompting.
Anthropic says Claude Code can now build dynamic workflows: task-specific JavaScript harnesses that spawn and coordinate subagents for complex work, a development that could change how teams use AI coding agents for large, parallel, or review-heavy tasks.
The feature, described by Anthropic researchers Thariq Shihipar and Sid Bidasaria in a June 2, 2026 Claude blog post, allows Claude to write orchestration code around itself. According to the company, that harness can route work, start subagents, wait for their outputs, merge results, and assign separate review roles.
In a July 1, 2026 analysis, Thorsten Meyer AI framed the feature as Claude drawing an organization chart for one job: a dispatcher, specialists, reviewers, and judges that are created for the task and then dismissed. That framing is the site’s interpretation, while the underlying mechanics and workflow patterns are attributed to Anthropic.
Anthropic’s caveat is central to the announcement: dynamic workflows are meant for complex, high-value tasks, not small edits. The company says the approach can consume meaningfully more tokens, because several agents may work in parallel or in sequence before a final result is produced.
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.
Claude Moves Beyond Solo Agents
The development matters because it targets a known weakness in agentic systems: a single agent can lose the original objective, stop early, or judge its own work too generously. Anthropic says dynamic workflows address those risks by giving subagents separate context windows, focused briefs, and, in some cases, independent verification roles.
For software teams, the clearest use cases are work that can be divided or checked: large migrations, major refactors, security reviews, research synthesis, ticket ranking, and root-cause analysis. If the approach works as described, users could ask Claude Code to coordinate a temporary team rather than push one overloaded agent through every step.
The tradeoff is cost and control. A workflow that spawns many agents can burn far more tokens than a normal session, and the quality of the result still depends on the task boundaries, stop conditions, and review criteria. Anthropic is presenting the feature as a manager-like pattern for hard work, not a default setting for every request.
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.
The Third Claude Code Pattern
Thorsten Meyer AI described dynamic workflows as the third part of a loose trilogy around Claude Code. In that framing, skills package organizational knowledge, loops decide how far to delegate over time, and dynamic workflows coordinate multiple agents within one task.
The article lists six workflow patterns that Anthropic says Claude can compose: classify-and-act, fan-out-and-synthesize, adversarial verification, generate-and-filter, tournament-style judging, and loop-until-done. The common idea is that Claude writes the control structure for the task instead of relying on a fixed prompt chain.
The security example is also specific: agents that read untrusted public content should be quarantined from high-privilege actions, with a separate agent assigned to act. That separation-of-duties model reflects broader concerns about prompt injection and tool misuse in autonomous systems.
“A harness for every task”
— Anthropic, via the Claude blog
JavaScript task orchestration tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Token Costs And Limits Remain Open
It is not yet clear how often dynamic workflows will outperform a well-prompted single agent in everyday developer work. The source material identifies likely use cases, but it does not provide independent benchmarks across cost, latency, accuracy, and failure recovery.
It is also unclear how teams should set practical guardrails. Thorsten Meyer AI warns that workflows can spawn many agents and use far more tokens, but the available material does not specify standard budget limits, default caps, or when Claude should refuse to expand a task further.
multi-agent AI system
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Teams Test Workflow Boundaries
The next step is adoption by developers and AI teams using Claude Code on tasks large enough to justify the added orchestration. Early evaluation is likely to focus on whether workflows improve review quality, parallel execution, and final synthesis enough to offset higher token use.
Readers should watch for Anthropic documentation updates at code.claude.com/docs, more detailed examples, and user reports comparing dynamic workflows with simpler agent setups. For now, the confirmed development is the capability itself; the broader performance gains remain a developing question.
AI review and coordination tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What are Claude dynamic workflows?
Dynamic workflows are task-specific orchestration programs that Claude Code can write to coordinate subagents, route work, merge outputs, and assign review roles.
Is this the same as using one Claude agent?
No. A single agent works inside one main thread, while a dynamic workflow can use multiple subagents with separate context windows and focused assignments.
When should teams use this feature?
Based on the source material, the best candidates are large refactors, deep research, security reviews, ticket triage, and other tasks that benefit from parallel work or independent checking.
What is the main downside?
The main downside is higher token use. Anthropic and Thorsten Meyer AI both frame the feature as suited to complex, high-value work, not small edits or routine requests.
What remains unproven?
The available material does not settle how often dynamic workflows beat simpler approaches on cost, speed, reliability, or final output quality in real production use.
Source: Thorsten Meyer AI
