The most disconnected people from AI are the ones credentialed to teach it

Over the past year I’ve come across three computer science deans and two professors from public and private universities — the people supposed to be running the departments that will mint the next decade of engineers. All of them waved off AI as a bubble. A scam. Granted, valuations are frothy, but that was not their point. In their esteemed opinion, it’s a fad, a passing thing serious people should not touch.

I will be precise about why they’re wrong, because reason is more useful than insult:

Let’s start with the word “bubble,” because it hides two claims pretending to be one. Is AI capital a bubble? Probably — valuations running ahead of revenue, half the seed decks a wrapper and a dream. That’s the financial observation and it’s fair. But “the money is frothy” tells you nothing about whether the tool works on a Tuesday.

Those are different ballgames. ChatGPT reached 100 million monthly users in two months. TikTok took nine. Instagram took two and a half years. Spotify, four and a half. UBS , after twenty years watching the internet, said they’d never seen anything ramp like it. It has more than a billion users now. The equity could be overpriced but it is still be true that no consumer technology in history has reached people this fast. A dean who can’t hold both ideas at once isn’t being skeptical. He’s just stopped learning, and is cling to a world that is moving on.

So why does the dean say it? It’s not his age — the field was built by people in their sixties who shipped more than I ever will. The answer is the incentive, and the incentive is structural. Tenure rewards publishing and being correct in print over a long horizon. Shipping rewards being wrong by Friday and fixing it by Monday. These are opposite metabolisms.

A man who has spent thirty years optimizing for not-being-wrong-on-paper is, by training, the worst-equipped person alive to evaluate a technology whose entire value is iterating through wrongness at speed. He isn’t protecting the truth. He’s protecting the identity he built before the truth changed. He hasn’t shipped anything a real user touched in decades, and he’s grading the people who will.

If you’re trying to cut out the noise and learn AI or software engineering, this part matters to you:

The academics and “experts” will tell you to learn algorithms and data structures first. Grind LeetCode. Earn the right to build by suffering bottom-up through the fundamentals. This was good advice in 1995 but it is malpractice now. Not because fundamentals don’t matter — because the order is backwards. And you are paying real money to be taught the backward one by people insulated from ever shipping the result.

Here’s the order that works better. Pick a real thing. Airbnb. Spotify. Amazon. Start at the top and ask what it does for the person using it. Then ask what systems have to exist for that to happen, and what each one earns for the business and the customer — search, ranking, payments, trust and safety, the recommendation layer. Then go one level down: what are the technical components inside each system. Then one more level, until you’re standing at the container, look inside, and understand why it’s there. Do that for three products and you’ll understand more about building software and data than most graduates with a four-year degree, because you’ll have learned the thing these degrees skip — how the pieces serve a purpose, top to bottom.

That’s the kind of engineer AI actually rewards. Not the one who can reverse a binary tree on a whiteboard. To understand a system well enough to direct a model through building it and to catch it when it’s wrong, is the whole game.

People say “garbage in, garbage out” like it’s an argument against AI. It’s the opposite. It means the human is still the variable. Bad operators get bad output. Good ones ship in weeks what used to take a team a year. The leverage isn’t the tool — it’s what you bring to it.

Fact: For thirty years the bottom 90% of every market — the independent restaurant, the neighborhood clinic, the family logistics firm, construction subcontractors — got nothing. The McKinseys and Accentures wouldn’t help them. A real transformation cost half a million just to assess and three million and three years to build, so these guys ran on duct tape and prayers, and watched the enterprise players pull further ahead.

That gap is now closeable by one competent person, or a small, nimble team. Operators who actually understand systems can walk into that business, charge tens of thousands for an assessment that is genuinely good, and then — instead of a three-million-dollar, three-year build — deliver a transformed operation for low six figures. Example: A working custom CRM, owned by the business and shaped to how it actually runs, instead of bending the whole company around Salesforce’s billing model. It’s the most under-served market on earth finally becoming reachable, and it’s reachable specifically because of the tool the dean is calling a fad.

So when someone with a title and a parking spot tells you AI is a bubble, hear what’s underneath it: the world rearranged itself faster than his identity could follow. You don’t have that problem yet, don’t let him give it to you.

If you can take Spotify apart, top down to the container, and explain why each layer exists, you’re already ahead of the curve. Bless you. If no, that’s the gap, and the man calling AI a bubble is the last person who will close it for you.

TL;DR: The people calling AI a fad are optimizing for an incentive that punishes being wrong fast — the one skill the technology runs on. Learn systems top-down, not algorithms bottom-up. Then go close the 90% of the market everyone left for dead.