What it is
Instead of a single agent doing everything, tasks are delegated to focused subagents (one researches, another reviews, another writes). Each one works in its own context and returns a result.
Why it matters to an agency
It splits big problems into manageable parts and improves quality: a reviewer subagent can audit what another one produced. It is a key pattern in Claude Code.
Example
A main agent delegates to an "explorer" subagent that searches the repo and returns only what is relevant, saving context.