The stock passed every test I had. That was the problem with my tests.

It came in the mail — a glossy newsletter, the kind that used to arrive in the early 2000s with a tear-off reply card, and I was young enough to read it to the end. The company had everything my checklist asked for: a sector I believed in, a chart that was going the right way, a story about a technology that was about to change an industry, a management team with the right words in their biographies. I ran my criteria. It cleared all of them. I bought it, and I held it with the special conviction of a man whose system has just confirmed his judgment.

The system had confirmed nothing. It had classified. It had checked whether the company carried the labels I was looking for, found that it did, and stopped. What it had never done — what I did not know how to do yet — was ask whether the thing behind the labels was real. Whether there was a business there, with customers and receipts, or just a company-shaped object with the right words on it. There wasn't. The position went to zero over a number of years, slowly enough that I could tell myself it was patience.

I think about that mailer more than I think about any of the companies I got right, because the error in it is the error of this entire decade, scaled up.

Classification is sorting by label. It's what a screener does, what an index does, what the market does on a busy day: everything carrying the word gets counted, and the counting is mistaken for knowing. Curation is the other thing — deciding what's actually there. It is slower, it requires reading, and it is the only part of investing that has ever been worth money.

The clearest illustration of the difference I've come across this year isn't in investing at all. It's a company you've probably never heard of, in San Francisco, with no venture capital, that earned more than a billion dollars last year doing one thing: it employs experts — by its own count, more than two hundred thousand of them, many with doctorates — to tell the largest AI labs in the world which of their models' answers are true. The labs can generate a million answers; they cannot, by themselves, tell the good ones from the plausible ones. So they pay a company of human curators to do it, and the curators have quietly become one of the most profitable businesses in the transformation. The entire AI economy, it turns out, runs on a layer whose only product is judgment.

That's the shape of it. Generation is cheap and getting cheaper. Classification is automatic. Curation — knowing what's real — is scarce, and the scarce thing is what gets paid.

Now carry that into the market. "AI" is the most valuable word in the economy right now, which means it is also the most freely attached. Every company that can plausibly say it is using AI says so; every fund that can call itself an AI fund has; every screener that filters for the term returns a list that grows by the month. A great deal of money is being invested by classification — buy what carries the label — and a great deal of that money is going to find out, over a number of years, that it bought the label.

The layers underneath don't carry the word on the door. The foundry that makes the chips, the company that sells the only machines that can etch them, the memory makers whose customers now pay cash deposits years ahead, the connectivity firms, the clouds, the enterprise plumbing — these are businesses with receipts, not stories with a sector. Some of them have been called boring for forty years. Their valuations rarely spike on the word, because the word was never their product. The product was always the layer, and the layer is what every label has to pay.

That is the distinction the whole structure of this letter rests on: owning the AI label is a classification; owning the AI layers is a curation. The first can be done in an afternoon with a screener. The second requires reading the business, deciding where value actually lands when the transformation is finished arguing with itself, and then — the hard part — holding the answer through the years when the label is outperforming it.

The hesitation sounds like this: How do I know I'm buying the real thing and not the label? I'm not a professional. I can't read a foundry's balance sheet.

Here's the calm response, and it's the one I wish I'd had with the mailer in my hand. You don't need to out-analyze the market. You need to ask one question, and it's a question anyone can ask: who gets paid when the label is wrong? When an AI application fails to earn its valuation, the chips it ran on were still bought. When a chatbot company's story collapses, the data center it rented was still built, the memory it used was still sold, the connectivity inside the rack was still shipped. The layers get paid whether or not the names on top succeed. That's not a sophisticated insight; it's the plumber's view of a building. And it has the advantage of being true in every past transformation — the people who sold rails and switches and routers did fine regardless of which railroad or portal or search engine won.

So the test isn't "does this company say AI." The test is "does this company get paid by the companies that say AI." Run it once, on a calm day, on a short list. Then let a schedule do the rest. The schedule's job is to keep you from re-running the test every time the label has a good week.

The company from the glossy newsletter would have passed any AI screen built today. Right sector, right story, right words. It had the label. It never had the layer — no one paid it to exist — and the market eventually noticed, slowly, the way it always does.

I don't run screens anymore. I read businesses, on calm days, and I ask who pays them. The list that survives that question is short, it's unglamorous, and it's the only list I've ever held through a full cycle without flinching. The transformation will produce thousands of companies that carry the word. It will be built by a few dozen that get paid by all of them. Curation is telling the two apart. Classification is what everyone else is doing.

— Christopher Cinek
Founder, AI Wealth Blueprint

Nothing in this publication constitutes financial, investment, tax, or legal advice. AI Wealth Blueprint is an educational newsletter published for informational purposes only under the "publisher's exemption" of the Investment Advisers Act of 1940. No personalized investment advice is provided. All examples, illustrations, and numeric scenarios are hypothetical and for educational purposes only. Past performance does not guarantee future results.