Scoping Agentic Retrieval vs Fixing the Basics

A senior engineer on your team wants to skip straight to an agentic, iterative retrieval loop (retrieve, assess sufficiency, retrieve again) for a RAG system that currently has: fixed-size chunks, a single dense index, top-5 retrieval, no re-ranking, no confidence gate. Their argument: "agentic retrieval is strictly more powerful, it can only help."

1. Is the argument correct that it "can only help"? Explain why or why not.
2. What would you recommend instead, and why, using the maturity ladder framing?
3. Describe one concrete query type where agentic/iterative retrieval genuinely earns its complexity, even after the basics are fixed.