Is Grep All You Need? How Agent Harnesses Reshape Agentic Search
Summary
The arXiv paper compares retrieval strategies (grep vs vector) in agent-powered search workflows using Chronos and provider CLIs, reporting that grep often yields higher accuracy in their experiments but emphasizing that performance hinges on harness and tool-calling style. It provides empirical results across two experiments and discusses how search context and noise affect retrieved results, offering practical insights for designing LLM-enabled agents.