Claude Code Dynamic Workflows

Claude Code dynamic workflows are a high-parallelism orchestration mode introduced alongside Claude Opus 4 8, where Claude generates an executable JavaScript workflow that spins up specialised sub-agents, coordinates their work, and merges the result back into the main Claude Code session.

Key points

• The workflow itself is generated dynamically from the user's goal; Nate describes Claude creating a JavaScript file, asking for explicit approval, then executing that script as the workflow plan ^[src-087].
• The mode is best suited to broad analysis, audits, scoring, or synthesis tasks where many independent agents can inspect different parts of a problem in parallel ^[src-087].
• Nate's skill-audit example created 41 Haiku scoring agents, one per skill, then fed their outputs into an Opus synthesis agent ^[src-087].
• Dynamic workflows overlap with Claude Code Skills, sub-agents, agent teams, and /goal, but the distinction is scale and structure: a generated workflow can coordinate a larger fan-out pattern than a normal single-session loop ^[src-087].
• The cost profile is input-heavy. Nate's example consumed about 5 million input tokens and another workflow prompt used roughly half of a $200 monthly subscription, so explicit approval and cost awareness are part of the operating pattern ^[src-087].
• Dynamic workflows are useful but not default. Nate says most automation and knowledge-work sessions do not need them; they are for large, parallel, expensive analysis jobs where the synthesis is worth the token burn ^[src-087].

Related entities

• Claude Code
• Claude Opus 4 8
• Anthropic
• Nate Herk

Related concepts

• Agent Teams
• Claude Code Skills
• Token Economics
• Agentic Workflows
• Self Checking Todo Loops

Source references

• ^[src-087] Nate Herk — "Claude Code Dynamic Workflows Clearly Explained" (2026-05-30)