This may be true for code bases. But for general knowledge bases that document a system using prose or natural language, the agentic search claims weaken. Especially when that knowledge base is composed of multiple/many documents (think enterprise scale).