An agent must assess the impact of changing refund logic in a large monorepo. Running every search and file read in the main session floods its context, and its coordination of the overall analysis is getting worse. What is the most effective design?

Answer: C — Spawn subagents for specific questions, like tracing refund-flow dependencies, and coordinate from their summaries

Delegating specific investigation questions to subagents isolates the verbose output while the main agent keeps high-level coordination. (A) moves the problem into the subagent and then hands the raw output back. (B) leaves the noise in context. (D) addresses output length, not context pressure.