2.4 Sub-agents

Info
⏱ Duration: ~10 min (informational) · 🎯 Goal: understand that sub-agents are native coordination Claude decides on, not something you launch manually.

Scaling doesn’t mean you orchestrate more pieces: it means Claude coordinates the work for you. When a task is complex, Cowork breaks it down and works several streams in parallel —without you launching anything.

Info
Reconciling with the basic track: in the basic workshop you modeled a “board of advisors” as a prompting technique —you wrote each voice in the message. Native sub-agents are different: Claude coordinates them on its own, decides when to fan work out, and runs them in parallel, without you writing or launching them.

Native coordination

Verbatim: “Claude breaks complex work into smaller tasks and coordinates parallel workstreams to complete them.” During a task, “for complex tasks, Claude may coordinate multiple sub-agents working simultaneously.”

  • You don’t launch them: Claude decides when to fan work out into sub-agents.
  • Cowork only: sub-agents (like hooks) run only in Cowork; they appear grayed out in chat.
  • Custom sub-agents arrive bundled in plugins —that’s the path to distribute and scale tailored sub-agents across your organization.
Tip
As a director, your job isn’t to micro-orchestrate sub-agents: it’s to delegate well and let Claude decide how to parallelize. Governance (modes, the Session 2.2 policy) is what you keep under control.

✅ Checkpoint

You can explain that sub-agents are native coordination decided by Claude, that they run only in Cowork, and that custom sub-agents are distributed bundled in plugins.

Got it? — 2 questions

1. Who decides when sub-agents are used? Claude: it breaks complex work down and coordinates parallel streams on its own; you don’t launch them manually.

2. How are custom sub-agents distributed across an organization? Bundled inside plugins —the same path used to distribute skills and connectors.

Workshop Claude Avanzado · Grupo Salinas · v1.0 · 2026