Users submit both broad questions ("How is climate policy affecting shipping costs?") and narrow factual ones ("When did the EU emissions trading scheme start covering maritime transport?"). Every query runs through all four subagents, so narrow questions take minutes and return a ten-page report. What is the best design change?

Answer: B — Design the coordinator to analyze each query's requirements and invoke only the subagents it needs, such as a single web search for a narrow fact and the full pipeline for broad topics.

A coordinator should analyze each query's requirements and choose dynamically which subagents to invoke rather than always running the full pipeline (task 1.2). A separate trained classifier (a) is over-engineered and needs labeled data when the coordinator can already make this decision. Shortening the report (c) hides the symptom while all the latency and cost remain. Keyword rules (d) are brittle: plenty of narrow questions don't start with those words, and some broad ones do.