RAG

RAG (Retrieval-Augmented Generation) is a technique for grounding language model outputs in external knowledge by retrieving relevant information at inference time. As framed by Kuba Rogut of Turbopuffer, RAG is broader than commonly understood: it encompasses not just vector search but also full-text search, grepping, globbing, regex, and filters — often used iteratively by an agent. 3:20

Scope and definition

Rogut emphasizes that retrieval is not a single operation but a family of techniques: "retrieval is not just vector search. It's a lot of different things." 3:20 This framing is important context for evaluating "RAG is dead" claims, which typically target only the simplest instantiation of the pattern.

Adoption trajectory

Despite social media narratives about RAG's obsolescence, Google search volume data contradicts them. Rogut notes that search interest hit a new inflection point in mid-2025 and "just goes through the roof," adding: "So take that Twitter." 1:53 This suggests mainstream adoption of RAG is still accelerating even as the practitioner frontier moves beyond it.

Points of disagreement

Rogut holds a nuanced, partly critical position on RAG:

• Simple RAG is outdated: The pattern of a single one-time vector search followed by generation "worked back in 2023, early 2024 at kind of the beginnings of AI" but sophisticated users have moved on. 14:57 He observes that "a lot of people are no longer doing the simple RAG." 14:57

• RAG as a category is not dead: The broader category of retrieval-augmented workflows remains very much alive and growing, as the search volume data illustrates. The title of Rogut's talk frames the contrast explicitly: "RAG is dead — how hybrid tool-rich retrieval is becoming a default for serious agentic search." 0:43

Relationship to agentic search

Agentic Search is positioned as the successor pattern to simple RAG for users seeking real performance gains. Rather than a single retrieval step, agentic search uses multiple retrieval tools iteratively, with an agent deciding which retrieval strategy to apply at each step. 0:43 The distinction between RAG and agentic search is one of the first clarifications Rogut makes in his talk. 0:40