Subagent coordination in Claude Code costs zero tokens

Developer 0xCodez posted a 14-step guide on X on July 20 for restructuring Claude Code’s dynamic workflows as a graph instead of a linear chain.
Dynamic workflows landed in research preview in late May 2026. Claude writes a JavaScript orchestration script, the runtime executes it in the background and spawns subagents from there. Intermediate results live in the script rather than the model’s context, so the coordination itself doesn’t consume tokens.
The core heuristic: at every ‘and then’ in your workflow, check whether the next step actually reads the previous step’s output. If it doesn’t, run those steps in parallel.
The guide is not a substitute for the official docs, and some details, including parallelism limits, conflict with other write-ups. The feature needs Claude Code v2.1.154 or newer and is toggled on Pro under Dynamic workflows in /config.
Up next
Whetstone: 20 Claude Code skills, each written after one real failure
Whetstone is a Show HN collection of 20 Claude Code skills, each derived from one real failure: a check that passed on a stub, a merge that dropped a field, a green test that couldn’t fail. Each ends in a checklist that fails loudly on a missed item.
Related stories
- Claude Code team defines a loop by its stop condition
- AGENTS.md request stays open on Claude Code tracker
- A Claude Code cron job now triages support tickets over MCP
- Claude Code stops wiping .env files
- Replit says its agent swarm built Radar in 2 days for $3,000
- Claude Code team has Claude test the app before committing
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.

