Showed it to my fellow Roman Petrochenkov yesterday — analytics agent on top of BigQuery, semantic layer, the works. You ask it a question in plain English, it goes and answers it.
He said something that stuck with me. Non-technical folks look at it and think it's magic — no idea what's inside, don't need to know. Devs look at it and get nervous, because they can see it's not magic, it's just config, and one bad prompt can break the whole thing.
Then we got to the actual point. Everyone's selling the same story right now: plug in an agent, it does the work, you sit back. That's not really what happens.
The agent is just a frontend. The real work is still the boring stuff — building a proper data model, making sure your metrics actually mean the same thing everywhere, cleaning up the mess before you put a chat window on top of it.
Get that right and you can reuse the same setup for client after client. Get it wrong and the agent will just lie to you confidently, in a nice UI.
He compared it to a boat with one hole in it — doesn't matter how nice everything else is, that boat isn't going anywhere until someone patches the hole. Most AI agents I've seen, the hole isn't in the model, it's in the data nobody cleaned up first.
So yeah, not really magic. Just plumbing that finally works.
Want a stronger data analyst role or a raise? Grab the FREE Product Analyst Playbook and get the exact roadmap to your next offer.
