A coordinator needs results from three independent search subagents, each covering a different subtopic. Logs show it emits one Task call, waits for the result, then emits the next, so each request takes about three times as long as a single search. What change best addresses the latency?

Answer: C — Have the coordinator emit all three Task tool calls in a single response so they run in parallel

Parallel subagents are spawned by emitting multiple Task tool calls in one coordinator response rather than across separate turns. Merging subtopics (a) gives up the scope partitioning, chaining subagents (b) is still sequential and bypasses the coordinator, and trimming tools (d) doesn't address the serialized turns.