One question — dozens of agents dig in parallel, the answer is assembled from the swarm.
A coordinator splits the task into slices, hands them to sub-agents, then merges their findings into a single answer — so the analysis reaches beyond a single context window, out of reach for a lone agent.