Member-only story
I thought retrieval-augmented generation was the easy part. The more I learn, the more I realize the tutorial version is hiding where the real difficulty lives.
I’m learning RAG properly right now — not the “watch one tutorial and feel like an expert” version, but actually sitting with it, building small things, and running into the gap between how simple it looks and how it behaves when you push on it.
This article is me writing down what I’ve figured out so far, specifically the things that surprised me. I’m not going to pretend I’ve shipped a battle-hardened system serving millions of users. I haven’t. What I can offer is something that’s weirdly useful: I still remember being confused by this, and I can point at the exact spots where the standard explanation quietly skips the hard part. If you’re learning RAG too, I think these are the things worth knowing early, before you waste a week learning them the slow way.
Let me start with the thing everyone gets told, because it’s where my confusion began.
The version of RAG everyone teaches #
RAG — retrieval-augmented generation — is how you get a language model to answer…