Workflows and agent teams

Claude Code has four ways to take on multiple tasks at once:

Mechanism	What it is	Best for
Subagents	Isolated workers inside your session	Focused tasks where only the result matters
Agent view	Dispatch and manage many sessions from one screen	Monitoring parallel sessions needing input
Agent teams	Independent sessions that message each other, sharing a task list	Work requiring discussion — competing hypotheses, parallel review
Dynamic workflows	A script Claude writes that orchestrates many subagents; rerunnable	Codebase audits, large migrations, cross-checked research

Dynamic workflows live in .claude/workflows/*.js — each file becomes a /<name> command. You don't author them from scratch: Claude writes them, and you save a run from /workflows with s.

Agent teams trade tokens for collaboration — each teammate is a separate Claude instance with its own context. Subagents report only to the caller; teammates message each other directly. If parallel subagents start hitting context limits or need to talk to each other, teams are the next step. (Experimental, disabled by default.)

Check your understanding

Question 1Subagent vs agent team — the architectural difference?

Question 2How is a dynamic workflow usually created?