Is RAG Really Dead? Agentic Grep and Hybrid Retrieval on Company Reports

Coding agents have moved from embeddings to plain text search, prompting claims that RAG is dead. We compared three configurations on 150 FinanceBench questions over 368 reports, using one open-weight model (qwen3.6-27b), with the main comparison on the same parsed markdown. Classic RAG searched once before answering. Agentic RAG and Agentic Grep shared an agent loop and could search repeatedly, with hybrid retrieval or grep respectively. Classic RAG answered 117 questions correctly, Agentic Grep 127, and Agentic RAG 133 and 132 over two runs (exact McNemar p = 0.21 and 0.33). Both agents about halved Classic RAG's wrong answers. The step limit mattered: in an earlier run at a ten-step cap, grep scored 121. Most FinanceBench questions name the company and period, letting grep narrow its search by file name. On 32 questions over a mixed library, many needing several documents, Agentic Grep scored 78%, Agentic RAG 84%, and VectoGrid, our product built on Agentic RAG with answer review and set handling, 91%. Our conclusion: Classic RAG is dead, but RAG is not. Agentic RAG answered 15-16 more questions correctly than Classic RAG, and stayed ahead of Agentic Grep throughout, most of all on the mixed library, though neither gap over grep is significant alone.