What is a sub-agent?
A sub-agent is an AI agent that a lead agent assigns part of a task to, with its own instructions and context. The lead agent writes a brief for the sub-agent, which works in a separate context window and returns a summary of its results. Hugging Face's agent glossary notes that a sub-agent can reason on its own, which a tool cannot.
Why it matters to a revenue leader
Sub-agents let AI take on bigger jobs, such as researching a whole list of target accounts, without one agent losing track of the details. They also cost more. Anthropic reported that its multi-agent setup used about 15 times more tokens than a chat, and found that vague briefs led sub-agents to duplicate work or leave gaps, the same failure a manager sees after an unclear handoff. If you pay usage-based AI bills, find out when a tool splits work across sub-agents and what each one is allowed to access.
Example
Fictitious numbersA utility services company asks an AI tool to profile 40 municipal accounts before budget season. The lead agent splits the list among four sub-agents, each researching 10 accounts, then assembles one brief per account. The research finishes in a fraction of the time, and the usage charge for the job is several times what a single agent would have cost.
"When our AI tools split work across sub-agents, what does each one see, and what does that cost us?"
Sources
- Create custom subagents | Claude Code Docs checked 9 Oct 2026
- How we built our multi-agent research system | Anthropic checked 9 Oct 2026
- Harness, Scaffold, and the AI Agent Terms Worth Getting Right | Hugging Face checked 9 Oct 2026