On 26 August 2026, Bill Gates published on his blog an essay of around 6,000 words, “The turbulent AI era is here. The choices we make now are critical.” (Gates Notes), proposing a tax on AI tokens and robots as well as the creation of jobs reserved for humans. If these proposals gain political ground, they directly change the profitability calculation of any automation undertaken today.
The facts
- On 26 August 2026, Bill Gates publishes an essay of around 6,000 words in which he observes that there is no plan to absorb the social, political and economic upheavals of AI (CNBC). He calls for national bodies able to set cross-cutting priorities across government departments, and for an international organisation inspired by nuclear inspections, civil aviation rules and the ozone layer treaties.
- He proposes taxing AI tokens and robots to rebalance taxation between labour and capital: hiring an employee generates employer contributions, while buying a robot is written off as an expense. The proceeds would fund retraining and the social safety net; Gates wants the tax targeted so as not to hold back beneficial uses such as healthcare or education, and accepts that it would not be economically optimal.
- He calls “Human Reserved” the jobs that machines could perform but that society would choose to keep for humans, on the model of a nature reserve: land that could be built on, deliberately preserved. In his interview with GeekWire, he says he explored with Anthropic's Claude chatbot scenarios that take the share of work reserved for humans up to 40%, by combining shorter working days and earlier retirement.
Why does this proposal shift the debate from technology to organisation?
Bill Gates's essay raises a question of arbitration that most management teams have yet to formalise: who, within the organisation, decides which functions AI will never replace. Two concrete mechanisms carry that question. The first, the tax on AI tokens and robots, would reintroduce on automation a charge comparable to the one that weighs today on employing a member of staff: a principle which, if it spreads, transforms the profitability calculation of every automation project under way. The second, “Human Reserved” jobs, assumes that an organisation chooses, upstream and as a matter of principle, the functions it refuses to hand over to AI, regardless of technical feasibility. Gates acknowledges it himself: who decides, on what criteria, how to prevent workarounds, all of that remains to be built in public.
What the two proposals have in common: they take away the comfort of deciding case by case. A management team that steers its AI adoption function by function, without a written decision grid, will find that choice imposed from outside if taxation or regulation moves in that direction, just as the AI Act already sets its deadlines for AI agents. One that has documented it beforehand, with criteria specific to its business instead of a line of communication, will negotiate its future compliance. The others will have to put up with it.
AIxH's view
Gates's proposal confirms a conviction we have held since the creation of AIxH's AI audit and integration practice: the decision to hand a task to AI, or to keep it, should never rest on technical feasibility alone. That is the heart of our work in AI audit and integration in Luxembourg: formalising precisely this kind of arbitration, which functions to automate, which to keep human, and why, before a future regulation or tax imposes it from outside. If those criteria do not yet exist in writing in your organisation, now is the time to set them down, while the choice is still yours.