Europe 2026
RAG is dead, right?? — Kuba Rogut, Turbopuffer
Overview
This talk challenges the notion that Retrieval Augmented Generation (RAG) is obsolete, arguing instead that hybrid, tool-rich retrieval is becoming the standard for sophisticated agentic search. While simple RAG, often equated with basic vector search, was effective in early AI development, more advanced applications now leverage agentic search. This approach involves agents iteratively reasoning over context using a suite of tools, including vector search, full-text search, and other filters, to progressively refine their understanding and complete tasks.
Who should watch
- AI engineers and builders exploring advanced search techniques for agents.
- Product Managers and designers evaluating the capabilities of agentic systems.
- Developers seeking to improve the accuracy and efficiency of AI-powered search.
- Those interested in the evolution beyond basic RAG implementations.
Key takeaways
- The perception of RAG being dead is often based on a narrow definition of RAG as solely vector search.
- True retrieval involves a combination of methods, including vector search, full-text search, and filtering.
- Agentic search is defined as empowering agents with tools to iteratively find and reason over context.
- Companies like Cursor demonstrate effective agentic search by indexing codebases using techniques like Merkle trees to efficiently update embeddings, leading to significant accuracy gains.
- Cloud Code, while not using vector search in its current iteration, previously experimented with RAG and found it less effective for their use case.
- Embeddings and semantic search can be viewed as a form of cached compute, reducing redundant processing and token usage for agents.
- Agentic retrieval is iterative, with agents performing multiple search steps and fetching only necessary information, moving beyond single-query RAG.
- Efficient retrieval mechanisms are crucial for large context windows, enabling agents to narrow down vast amounts of data to the most relevant subsets.
Notable quotes
*You don't need a trillion at once, you need the right million.*
Unofficial community note. Prefer the recording for nuance.