Ask someone who listens to dance music whether they listen to EDM. Watch their face contort into a funny shape.
EDM isn't a genre. It's a shelf with records on it. House is on it, and house splits into deep, tech, progressive, Chicago, French, and about nine more. Dubstep is on it, and Burial and Skrillex are not making the same record, even though they’re both technically dubstep. Drum and bass is on it. Electro, minimal, Japanese electro --they’re all on the shelf. When somebody says "EDM," they've named the shelf and told you nothing about the records.
"AI" is the same shelf.
What's actually on it
When someone says their product uses AI, ask the follow-up.
Is it a large language model? Then you're talking about transformers, attention, tokenization, context windows, and a very expensive guess about what token comes next.
Is retrieval bolted on? Then you're talking about a vector database, and now there's a real question: A purpose-built vector store, or Postgres with pgvector on top? Those are different systems with different failure modes, different cost curves, and different people you call at 2am.
Is it looking at something? Which thing, specifically? Image recognition is not image generation. OCR is neither. Audio recognition is not video recognition. Four different problems, four different architectures, but people say "computer vision" like it's one button.
And there's a whole category that never makes it onto the shelf at all. Fourier neural operators, like the work behind Nvidia's Darcy, predict physical systems by learning operators over function spaces. Weather models. That's an intelligent system. It will never write your email, and it is better at what it does than anything that will.
The part the marketing skips
I build deterministic systems. Causal ones. Systems that model why something happened instead of predicting what usually comes next. Posets, which are directed acyclic graphs of events ordered by cause, and an actual source of truth you can point at.
That counts as intelligent. It's what a person does walking into a room and working out what happened: They trace causes backward. It just doesn't happen inside a neural net, so nobody gives it the sticker. But I still think it should count.
If a system can't tell you why it produced an answer, you don't have an answer. You have a loose suggestion, supplemented by a model.
So here's the deal
I'm not teaching you AI.
I'm teaching you the technology underneath it, and that hasn't changed. It has evolved, but it’s still the same raw stuff.
There's still a database, and it still needs a defined shape. There's still a front end, and it still has to look good on a phone. There's still an API, and it still needs auth, rate limits, and versioning. Input still gets validated. Secrets still stay out of the repo. Somebody still owns the pager.
The new tools are real. Sometimes they turn a week into an afternoon, and I use them every day for exactly that. But that afternoon still ships to a server that has to be reliable, and not a massive target for black hat dickheads. The rules of computing didn't change. The way we program just changed. It’s faster. It’s easier. It’s also more error prone.
With speed, sometimes there are tradeoffs.
Who this is for
A career switcher with no background. Self-taught and stuck somewhere in the intermediate swamp. Bounced off a bootcamp and figured the problem was you.
I make the same curriculum for all three, because the fundamentals don't care where you started. You'll learn Python, SQL, Docker, and how data actually moves through a system. Then you’ll learn how to make that system move with alarming speed. Then when someone tells you their thing is powered by AI, you'll know which question to ask.