How agents work · Subagents
Subagents¶
A subagent is a helper AI that your main AI hands a task to — like a lead contractor bringing in a specialist. Useful sometimes. Not needed nearly as often as the internet suggests.
When they actually help¶
- Big reading jobs. "Have helpers each read one report; you combine the findings."
- Checking work. One helper writes, a separate one verifies. The writer never grades its own homework.
- Different hats. A "researcher" and a "skeptic" reviewing the same claim.
When to skip them¶
If one AI can do the job in one pass — which covers most everyday work — subagents just add cost, time, and coordination confusion. One agent doing one clear job beats three agents doing a vague one.
How to ask (no setup required)¶
In tools that support it (like Claude Code), you can simply say:
Copy this
Use one subagent to gather the evidence and a separate subagent
to double-check the three most important claims. You write the
final answer only after reviewing both. The researcher may not
approve its own claims.
The rule that keeps it sane¶
Subagents follow the same rules as the main agent: stay in the folder, plan first, you review the result. More agents never means less supervision.
If this helped
If this saved you from an over-built setup, a small tip on Ko-fi keeps every guide here free. Tip on Ko-fi