The main problem isn’t embeddings, in my experience, it’s that “vector search” is the wrong conceptual framework to think about the problem We need to think ab

The main problem isn’t embeddings, in my experience, it’s that “vector search” is the wrong conceptual framework to think about the problem We need to think about query+content understanding before deciding a sub problem happens to be helped by embeddings. RAG naively looks like a question answering “passage retrieval” problem, when in reality it’s more structured retrieval than we first assume (and LLMs can learn how to use more structured approaches to explore data much better now than in 2022) https://softwaredoug.com/blog/2025/12/09/rag-users-want-affordances-not-vectors https://softwaredoug.com/blog/2025/12/09/rag-users-want-affo...