Tech and AI
A proposal to put some jobs formally out of reach of machines, and to tax the machines that take the rest
By Staff Writer | 27 August 2026
Bill Gates published a long essay on 26 August arguing that governments have no plan for what artificial intelligence does to employment. His two concrete proposals are a category of work reserved to people by decision rather than by capability, and a tax on tokens and robots to remove the cost advantage of replacing a worker.
The essay runs to roughly six thousand words and is titled "The turbulent AI era is here. The choices we make now are critical." It opens by saying that a plan is needed to ensure the good outweighs the bad, and it spends most of its length on three things: the loss of entry level work, the arming of people who mean harm, and the weakening of social and cognitive development in people who stop practising either. Its recurring argument is that none of that can be handled by one country or one institution acting alone.
I believe that as AI and robots improve, we'll set aside certain things for only people to do.
Bill Gates, writing in his own essay of 26 August
Human Reserved
That is the name he gives the category, and the analogy he draws is to a nature reserve: land that could perfectly well be developed, left alone because what would be lost is worth more than what would be gained. Applied to work, it means a job stays with a person after the machine can do it more cheaply, because society has decided it should.
He does not pretend the idea is worked out. In the essay he sets out the questions he cannot answer: who decides what is reserved, what criteria are used, how a company is stopped from quietly using robots anyway, and what happens to trade when one country permits automation of something and another does not. In an interview published alongside the essay he offered child care and jury service as examples of work people are better suited to, and said that in education and health care the answer is not a ban on the technology but protection for the profession while its members use it.
He puts an upper bound on it himself. In what he described as a very extreme version, he could imagine 40 per cent of jobs reserved to people at the outset, and said that is as high as he could get the number.
The tax argument, which is the sharper one
The second proposal is narrower and easier to legislate. He argues that a company pays payroll taxes on an employee and pays nothing equivalent on the machine that replaces them, so the tax code itself tilts the decision. Taxing tokens and robots would remove that tilt. In the interview he described the change he is asking for as larger than any alteration to the tax system in his lifetime, arriving at a moment when politics is more divided than at any point in it, which is a fair account of the difficulty.
He also proposes a new international body for artificial intelligence, and compares the shape of it to weapons inspection regimes, aviation regulation and the treaties that protected the ozone layer, saying a new organisation would need elements of all three and more.
Why this lands differently from the usual warning
Plenty of people in the industry have warned about their own products. What is different here is the direction of the remedy. Most proposals accept that the technology arrives and argue about retraining, safety nets and transition funds. This one says retraining will not be enough, and that the answer is to decide in advance that certain work is not available for automation at all, then defend that line in law and in trade policy.
For anyone running a business that bills for professional time, the practical question is not whether such a reserve gets legislated. It is whether procurement rules, public sector contracts and professional regulation start writing in a human decision maker as a condition, which is a much smaller step and one that a client can take on its own. That is where this argument will be felt first, long before any tax on a token.