The Day I Got RAGGED

I asked Gully for a coffee last Thursday. I wanted to show off what I knew.

He runs AI infrastructure for some of the most respected firms in the UAE. Law firms. DIFC. Government-level security. The kind of work where the agent has to be bulletproof because the lawyer signing the email cannot be wrong. Two years he has been at it. Real systems, in production, paid for.

I run a small AI-marketing consultancy out of Dubai. Two months old. I had a lot of stories I thought he might find interesting.

We sat down. About twenty minutes in, he asked me if I knew about RAG.

“Of course,” I said.

I did not.

I had read the letters. I had seen them in posts on LinkedIn from people who looked like they were typing them with one hand on a vape. I had assumed it was the next thing after MCP, which I had also nodded about for the first six weeks of knowing about it.

Gully is a kind man. He must have spotted the lie inside the first sentence I tried to construct after my “of course.” He explained it the way a senior consultant explains a complex deal to a junior who is going to have to write it up later. Slowly. Patiently. Without ever making me feel stupid, even though I was very obviously the dim one in the conversation.

The penny dropped about forty minutes in. RAG — retrieval-augmented generation — is the difference between asking ChatGPT a question and asking an agent that has been given the right paragraph from the right document at the moment of being asked. It is the difference between a clever parrot and an actual specialist.

Most “AI agents” on the market right now are clever parrots. Most of mine are clever parrots. Gully’s are specialists. The gap is not vocabulary. The gap is two years of work on the plumbing.

This is not a comfortable thing to write. I ran the IBM account at Ogilvy. I generated six figures a month at Clicksco. I have built hundreds of websites. Today I spend a million dollars a year of clients’ money in the insurance space. I am not used to being the junior in any room. I am used to being the one who knows things.

For two months I have been using AI like a sharp marketing tool. I have shipped real work. Blog posts. Decks. Websites. Campaigns. Faster than I could have alone, often better. I had begun to think I understood what AI was.

Gully showed me, in forty minutes, that I had been playing with the toys in the front shop. The factory is somewhere else.

The factory is full of people who know what RAG means. Vector databases. Embeddings. Hybrid retrieval. Re-ranking. Admission rubrics. Cost gates. Half a vocabulary I did not have until last Thursday. And it is not vocabulary in the way “ecosystem” and “synergy” are vocabulary. It is description of how the actual systems work, by the people who actually build them, for clients who actually pay for them.

There is a stratification happening in AI right now that the market has not yet processed. There are people who use AI in chat windows. There are people who build with AI as infrastructure. The gap between the two is widening monthly, and the second group is not very loud about being in it.

I came home and asked Claude to teach me what RAG actually was.

That was the next day. That is the next post.