The “AI is here to stay” crowd will evaporate as soon as they have to start paying.

And it goes further than that. Imagine if the hyperscalers actually succeed at their goals. They create truly useful agent models. But crucially to their profits, these models can only be built at scale. Their business model fails if someone learns to duplicate their work on sane levels of compute. So let’s say everything goes OpenAI’s way. They invent truly useful models, and the tech can only work at the massive scale they’ve invested in, so they don’t have to worry about being cut off at the knees.

Now imagine you’re a knowledge worker. A programmer. An engineer. An analyst. An editor. Really any job that is done sitting at a keyboard, manipulating data in one form or another. Now imagine you’re a knowledge worker and you adopt these new models. You become dependent on them. Less and less of your technical skills actually reside in your own mind. The difference between your knowledge base and that of any rando off the street is now less and less. At some point, you’re completely dependent on them to do the most basic functions of your job. Why shouldn’t OpenAI charge you or your employer a license fee equal to half your salary? Why shouldn’t your boss respond by cutting your salary in half and paying for the LLM license that way?

If the great fever dreams of the hyperscalers comes true, basically every white collar employee in the country sees their labor bargaining power collapse. It would be like the decline of weaving as a profession in England at the start of the Industrial Revolution, which spawned the historical Luddite movement. What was once a skilled profession requiring years of formal apprenticeship became mechanized low-skilled labor that could be (and often was) done by literal children. In OpenAI’s ideal future, that is what will happen to anyone that currently makes a living working at a keyboard.

It would devalue labor by making it less specialized. The real skill still present would be those who are just good at carefully stating prompts to the god machine. And if you’re good at stating prompts towards one end, you’re likely good at stating prompts across many domains. If we really had the kind of LLMs that these companies dream of creating? Prompt engineering would become a mandatory class in high school. You wouldn’t be able to graduate high school without learning how to use them. It would be as critical a skill as writing. But it would be a skill that everyone possessed, and thus of little ability to command a decent living from.