What is a Sub-Agent?

Sub-agents exist to manage context and specialization. A single agent trying to hold an entire complex task in one context window accumulates noise — intermediate tool output, exploratory dead ends — that crowds out the reasoning that actually matters. Delegating a well-scoped piece of work to a sub-agent lets that exploration happen in an isolated context, with only the distilled result returned to the parent, keeping the orchestrator's own context focused on synthesis and decision-making rather than raw search output. Sub-agents also enable parallelism: independent pieces of a task — researching different questions, reviewing different files, verifying different claims — can run concurrently as separate sub-agents rather than sequentially in one thread, cutting wall-clock time for decomposable work. The tradeoff is coordination overhead: a sub-agent that starts with zero shared context needs a self-contained brief, and the parent must synthesize multiple sub-agent reports rather than trusting one continuous train of thought.