Subagents

A subagent is an AI agent that a parent agent, often called the orchestrator, starts or calls to handle one bounded part of a larger task, usually with its own instructions, its own tool set and a fresh context window. The parent passes a task description, the subagent works through it in its own loop, and only a condensed result comes back into the parent's context. A peer agent in a multi-agent system has a standing role and goals of its own, while a subagent exists only for the duration of the delegated task and reports to a single caller. Subagents appear in coding agents such as Claude Code, in deep-agent frameworks and in orchestrator-worker architectures, where isolating context keeps the main agent's window small.

The main reason to use subagents is context management: searching a large codebase or document set can fill tens of thousands of tokens, and delegating that search keeps the noise out of the parent's reasoning. The cost is extra model calls, added latency and lost detail, because the parent sees only the summary. Teams usually give subagents a narrow tool set, for example read-only search, and a fixed output format, so the parent can rely on what comes back.

Work with us

Ready to put agentic AI to work?

Book a free 30-minute consultation. We'll map one real process worth automating with production-grade AI.