A barista who'd made coffee for two years and one who'd made it for ten used to be worth different money. That gap was the whole game. Then the 2025 Brynjolfsson, Li and Raymond study in the Quarterly Journal of Economics put a generative-AI assistant on the desks of customer-support agents and watched what happened. Average productivity went up 14 percent. But the least experienced workers jumped 34 percent, and the best performers barely moved.
Read that again. The machine dragged the rookies up to nearly where the veterans already sat. It handed out the thing the veterans spent a decade earning, for free, on day one.
If you're the rookie, thats a gift. If you're the veteran, your ten years just got a lot cheaper, because the thing that made you worth more is now sitting in a chatbot anyone can open.
This is the whole story of value and nobody wants to say it plain. A thing is worth something because not everyone can do it. A good chef gets paid and a microwave doesn't. That's scarcity doing the work. The second everybody can produce the same output, the price of that output falls through the floor. Generic AI is the great flattener. It takes skill that used to be rare and makes it common, and common is worth close to nothing.
So every owner faces one real question. Using AI is a given, you'd be mad not to. It's what you're building that stays scarce while everyone else drinks from the same well.
David Autor at MIT ran the numbers in another 2024 QJE paper and found most of the jobs people do today are "new work", roles that didn't exist before 1940 and got invented since. Value doesn't sit still. It keeps walking off toward whatever's new and rare, and it leaves the commoditised stuff behind like an empty servo on the highway. The lifting jobs went. The thinking-in-a-box jobs are going now.
The generic model everyone shares is that empty servo. Your competitor has the exact same ChatGPT you do, same Claude, typing the same prompts and getting the same beige answers back. There's no moat in a tool the whole street already owns.
The moat is the thing that's yours and only yours. Your fifteen years of quotes. The reason you knocked back that one client in 2019 and how right you turned out to be. The way your best estimator prices a tricky job by feel, the stuff he can't even explain. That's your business memory, and no model on earth has it unless you feed it in. An AI running on your own data, trained on how your shop actually thinks, is scarce by definition. It can't be copied because the inputs can't be copied. Someone can clone your website in an afternoon. They can't clone fifteen years of you being right about your own trade.
Everyone's about to have the same generic brain bolted onto their business. Same speed, same slightly-too-smooth emails. When everybody has that, having it is worth nothing, same as the rookie support agents who all got dragged to the same middle.
What you know, held somewhere only you can reach, is the last thing that stays rare. Build the AI that remembers your business, not the one a million other people are already renting.