Recursive agents apply an agent pattern at more than one level of a task. A parent can create or call child agents that receive narrower objectives, return results and may themselves divide the work again.
Recursion can scale decomposition and parallel work, but it can also multiply cost, duplicate effort and lose context across levels. Implementations need depth limits, shared budgets, clear result contracts and a way to stop repeated delegation that is not producing progress.
ELI5
A recursive agent solves a large job by giving smaller parts to another instance of the same kind of agent. Those smaller agents can divide their work again when needed.
For example, a research agent might assign separate topic sections to child agents and combine their findings. The system needs limits so the children do not keep creating more agents forever or repeat the same work without adding useful evidence.
