If you look at the totality of all jobs, you will realize that most people have jobs in which they manipulate physical objects in some form as core function of their job. Take the stereotypical entry level / student job of flipping burgers. ChatGPT can't flip a burger. You would need an AI-powered robot to flip that burger. And between the cost for the robot and the cost for the AI, it isn't obvious that the minimum wage human doing the job isn't actually a lot cheaper than the robot. Also, even an entry-level human employee tends to do better when things go wrong. Imagine the flat-top griddle having an electrical failure and not heating anymore. A human would notice instantly that his burgers aren't sizzling anymore and react in some way. A robot might happily keep flipping raw burgers, with catastrophic consequences when that raw burger reaches the customer.

In the absence of physical object manipulation requiring a robot, from the remaining jobs there is still a very large part that requires direct interaction with humans, typically customers. You don't need a robot for that, but you need your AI to not mess up by hallucinating. There are famous examples, like the Air Canada chatbot inventing a refund policy, and the airline being forced by the courts to pay that refund. Basically the courts found that a company is liable for their employees messing up, regardless of whether that employee is human or AI. That can get expensive for a company, when a tech-savvy customer persuades your chatbot to sell him a $70,000 car for $1, an error a human salesman is unlikely to make.

If I use the version of Google Gemini that is included in a regular Google search, or I use a regular ChatGPT chatbot, that use of AI is free. That is the typical tech business model of reaching a dominant market position by providing a service at a loss. Which is inevitably followed by some form of "enshittification", because sooner or later the company needs to provide its services at a profit to themselves. Those trillion dollar AI data center investments don't pay for themselves if you don't charge your customers sufficiently. AI companies already charge "pro" users and commercial users, and some companies had to put caps on AI use, because AI use quickly got too expensive. So even for the jobs that an AI can do well, the question is how much cheaper it is going to be than a human, and how the quality and reliability of the output compares in the end.

Sam Altman is predicting the Billion-Dollar Company of One, a startup built by one person with a laptop, an internet connection, and an army of AI agents, worth a billion dollars. While I don't think that is impossible, it is obvious that there are huge parts of the real economy in which such a company is unthinkable. The company can't be involved in any manufacturing, can't be involved in any handling or delivering of physical objects, and can't be involved with any services to humans that can't be done purely online. Robo bricklayers? Robo hairdressers? Robo plumbers? All of these are still decades away even in the most AI-positive scenario, and might never become a practical reality out of cost reasons. I am absolutely positive that AI will not take *all* of our jobs. Although we might end up with some surprises on which jobs AI eventually will replace; for example I can envision a future in which much of porn is AI generated, putting a lot of pornstars out of work. But if you look at a general classification of occupations, you immediately see a huge number of jobs that AI is extremely unlikely to perform in the foreseeable future.