In a recent essay on artificial intelligence and work, Bill Gates argues that we should intentionally keep some jobs for humans, so as to preserve human work.
The essay uses the term “Human Reserved” because it suggests a nature reserve. Like a place we could build roads in, and don’t, because the loss would be too great. The essay unimaginatively cites case studies such as elder care, some teaching and health care (one example being the delivery of a terminal diagnosis). This sounds as awful as Bill Winters’ “low value human capital”. A nature reserve is an area set aside by an authority for the preservation of something quaint and archaic, to save it from processes that would change it. I love nature reserves, of course, and want more of them. But I don’t think people are like that. And when applied to labour the metaphor means the transfer of all agency to someone or some body who or which decides what is human work. Gates’ examples are scarily familiar and therefore so narrow as to be terrifying as to his views on the rest of work.
My concern is that the essay by Gates shows that the tech guys are not even ashamed to see human endeavour relevant only to its most basic categories, and where it might suit them. Even with good will to the author this smacks of paternalistic planning, wrapped up in conservationist language. So let’s look at this proposal on its own terms, say from the point of view of classical economics. First, a tradition that Gates’ tech industry has long professed to revere: Friedrich Hayek. Hayek’s main objection to economic planning was not the stupidity or ill will of planners, though both elements are possible but not necessary in central planning. Rather he argued that planners could not possibly have the knowledge needed to plan well. The relevant knowledge — of what a particular person’s time is worth to them, what trade-offs they are prepared to make, what they would rather be doing — is diffuse and ever-changing. No matter how well-intentioned, no board can know in advance whether a fifty-five-year-old former economist would be better off in elder care or construction or doing something that no one on the committee thought to protect. And also, note that, contrary to what Martin Sandbu said in his FT article on the “socialist calculation debate” on 29 August 2026, Hayek did not think this could be remedied by more computing power. In Hayek’s view, knowledge actually emerges and evolves from and with the interaction of people with the market in a spontaneous fashion that no amount of computing can replicate. Gates admits this problem in his own essay when he writes that "you can't tell" that worker where to go and "expect them to find it fulfilling" — yet the solution he offers requires exactly the kind of advance determination that won't work.
We have to understand that a truly free market is not just a market without a reserve list; it is a condition in which the participants have real choices. Hayek himself was explicit that freedom of choice necessitates more than one option. If we are confronted by a monopolist, he wrote, we are completely at his mercy. A single planner controlling the entire economic system would be the most powerful monopolist imaginable. This is the risk that AI poses. If a few firms end up in control of the compute, the models, and the terms under which labour can compete with or cooperate with them, then this is not a free market for labour at all — this is monopsony in market drag. To be able to make the Hayekian case, there must be real choice — i.e. labour, humans, must have real decision-making power (say, bargaining power), and for that, there must be real options, otherwise the "market" is not a market in any meaningful sense.
This is where the work of Adam Smith and David Ricardo provide some additional tools to think this through. Smith warned against the "vile maxim of the masters of mankind" — the tendency of those who have concentrated economic power to arrange things to their advantage and then to call the result the natural order. A committee deciding which trades humans may keep is a variant of the very stitch-up Smith spent The Wealth of Nations attacking, in guild restrictions and merchant conspiracies against the public. Smith’s alternative was a system of natural liberty, in which no faction — landlords, manufacturers or, now, platforms and AI labs — gets to set the terms for everyone else. David Ricardo helps us think about what to do with the returns AI creates, and thus how to ensure there is a truly free market. Much of the value flowing into AI companies today is not like normal profit, earned through risk and productive contribution. More and more, slices of the returns look like rent, Ricardo’s term for a surplus that accrues to whomever controls a scarce, differentially advantaged factor, regardless of any sacrifice made on his or her part. Land was fertile or well-situated without effort on the part of the landlord who owned it. Frontier AI models are powerful largely because they are trained on the accumulated corpus of human writing, art and labour. The chips are controlled by Nvidia, and they have a self-financing model for others to keep buying. And the growing dependence on cloud providers, those who own the computing resources, starts to look also like rent in the Ricardo sense, and not a return to entrepreneurial risk.
So Gates’ "Reservation of tasks" vs an unchecked "hands-off laissez-faire market" is a false trade-off. Rejecting the idea of job reservation is easy enough based on the kind of economics that tech entrepreneurs normally cite. But that doesn’t mean doing nothing.
To move forward with AI we should recognise the unique value of human labour, both in creating goods and services, and in making decisions about the relevant trade-offs between work and leisure, consumption and investment. Indeed, in a market, only humans have something to give up, in terms of time, so this must be the deciding factor in the relevant trade-offs.
In order to ensure that market makers, that is humans, are able to direct value, then we need to think through how to place human value at the centre of the economy. A start would be to capture the rent from AI, for example through taxation of AI’s economic surplus, structured the way Ricardo would recognise, not the mild write-off adjustment Gates proposes. This can be done in a way that does not undermine the incentive to innovate and move forward with AI infrastructure. This money should be used to fund labour’s (let’s call it humans’) actual bargaining position: income support, retraining, and the freedom to walk away from bad terms, which is what makes any market a market at all.
Let the division of labour move freely, as Smith and Hayek both insisted it must. Let AI compete with labour, allowing people to use it, reject it, work with or against it. But make those choices real by ensuring humans have the power to choose. And refuse to let those who control AI’s rentier returns also decide, by committee, what the rest of us are for.