A concrete example
My main agent writes the article. For the research, it delegates to an Explore subagent. For the review, to a Code Review subagent.
Why it matters
A subagent keeps a clean context and a sharp specialty. You stop your main agent from mixing everything together.
You run into it in Codex, Claude Code, research workflows, reviews, and multi-agent curation.
Don't mix it up with
Agentic workflow: An agentic workflow is a sequence of steps where the AI plans, uses tools, checks its work, and keeps going until it hits the result.
Multi-agent: A multi-agent system puts several specialized agents to work on the same mission.
Common mistakes
- Spinning up a subagent for every little thing and losing the benefit.
- Forgetting to define the subagent's mission clearly.
- Not passing it the right info as input.
Quick checklist
- First I check whether the word names a concept, a tool, a risk, or a metric.
- I tie it to a concrete case: My main agent writes the article. For the research, it delegates to an Explore subagent. For the review, to a Code Review subagent.
- I keep the main trap in mind: Spinning up a subagent for every little thing and losing the benefit.
Quick questions
What is Subagent in AI?
A subagent is a specialized assistant I hand a specific slice of the work to.
Where will I run into Subagent?
You run into it in Codex, Claude Code, research workflows, reviews, and multi-agent curation.
Which word should I read next?
Start with Agentic workflow, Multi-agent, Skill & Agent.