The Human Reserve

A reserve is what we build for a species that has stopped being able to compete. Every reserve has a warden - and Gates does not say who.

Bill Gates chose the metaphor himself. In his latest 6,000-word Gates Notes essay he proposes that certain occupations be set aside as “Human Reserved”. He says the phrase pleases him as it reminds him of nature reserves, “places where we could put buildings and roads but we choose not to because the loss would be too great.”

A reserve is what we build for a species that has stopped being able to compete. It has a fence, a warden and a title deed. The animals inside don’t hold any of these. Gates offers the image as an act of care. It also pre-emptively concedes the entire argument.

The diagnosis - that governments are unprepared - is sound. And he is more candid than some of his tech industry peers. His strongest case deserves stating too: some work is human because its being human is the point. An automated jury is not a jury. The instinct is a good one, and it is clear society will seek some form of protection for workers who cannot transition to new roles. He is sketching a direction, not drafting a bill; but a direction is tested by imagining the bill. And it is the instrument, stretched beyond those cases, that fails.

It protects a unit that no longer exists. A job is a legal wrapper around a bundle of tasks - the US Labor Department’s O*NET database unbundles a thousand occupations into over 18,000 tasks - and AI arrives one task at a time: the deconstruction we discussed in Judgement. Exposed. - Porter’s value chain runs through work itself, atoms recombining wherever they are done best. Gates supplies the test case. There is no technical reason, he writes, why a robot could not tell you that your disease is incurable, “yet it shouldn’t”. Reserve the doctor, then, and what does the statute actually leave still standing? The AI model reads the scan, weighs the clinical trial evidence, and drafts the letter to your GP. What remains reserved is the delivery of the news. A profession narrowed to a bedside manner, and no regulator could say when it happened. Who carries the liability when the model errs is a question the tech industry prefers not to raise; we took it up in the same essay. Is an automated jury still a jury, and is an automated diagnosis still a diagnosis? The first question is one of legitimacy, the second of accuracy. Different clinical circumstances will no doubt produce a hybrid variety of patient-doctor-AI services.

Who would the statute even bind? Only the firms that already employ. WhatsApp reached 450 million users with 55 staff, Instagram sold for a billion dollars with thirteen, and the disruptor never dismisses a protected worker, having never hired one. The reserve fences the payrolls of 2026 while the employers of 2032 build AI-native startups outside the Human Reserve’s perimeter wire. Draft the statute around the activity instead, as licensing is drafted, and the doctor returns: the tasks migrate until the licensed human is a signature.

It regulates the wrong side of the race - and if the social case leaves you cold, it also shrinks the consumer market and the tax base every statute leans on. A hiring statute leaves the technology’s pace untouched; what slows is human adjustment. In 1900 four in ten Americans worked on farms, against fewer than two in a hundred now, and a government that had reserved farm work would have kept its people in the fields while the mills and the century went elsewhere. Such transitions are cruel - in Britain’s, real wages stalled for two generations, the stretch economists call the Engels pause - but cruelty argues for speeding people through, not for holding them at the gate. American railroads ran the controlled experiment: a fireman on every diesel locomotive from 1950, paid daily to tend a fire he knew was not there, until arbitration abolished 18,000 of the posts in 1963 - though every man with more than two years’ service kept his employment rights to retirement. That is reservation at its best: a pension for those inside, a closed gate for the next man. The practice was called featherbedding, and its lesson concerns the man as much as the money. Work known to be ornamental corrodes the worker long before it costs the employer, and years inside the fence are years not spent adapting.

Nor can the reserved economy stop trading with the unreserved one. Strip the metaphor and the policy family is an old one - protectionism, applied to labour. America’s dockworkers won contractual caps on port automation in 2025. In the World Bank’s latest performance index, published this June, four Chinese terminals and Salalah hold the top five places; the only American port anywhere in the world’s top 50 is Philadelphia, at 48th. Porter’s observation about nations holds: sheltering a factor of production from competition exports the industry it means to preserve.

And why should the customer cooperate? Waymo reports 220 million autonomous miles with 94% fewer serious-injury crashes than the human benchmark, figures which, once they travel from mapped cities to freight, will make a human-reserved lorry driver look less like a protected worker than a legislated hazard. Wherever consistency, speed or reliability is the product, people will demand the machine standard - the reserved workers included, the moment they are the customer. What stays human drifts toward the performance of humanity itself: craft, provenance, artisan bread priced by its unevenness - a fine niche, and as national labour policy a theme park. Or we fill the hours with games - physical, digital, social. Juvenal had a name for keeping a population fed and diverted once its labour no longer mattered: bread and circuses.

Even the statute’s timing is wrong. Sholto Douglas of Anthropic expects models able to automate most computer-facing work by 2028, yet people working those jobs into the 2030s, held there by compute shortages, diffusion friction and unmet demand: a human reserve already exists, made of friction and regulated roles - and friction decays on its own schedule. The years it buys will be spent adapting or behind the fence - and even that is first-order thinking against the question Gates never reaches, of who owns the reserve.

Silicon Valley spent this month arguing, Gavin Baker against Anthropic’s researchers, over whether increasingly powerful AI models are too dangerous to concentrate in the hands of a few, or too dangerous to distribute across the population at large; Douglas’s own concession, that inequality of compute will matter more than open models, is the sharper point. Abundance flows to whoever holds the intelligence, and a population whose work is permitted rather than demanded has lost its bargaining position. Serfdom, too, came with formal guarantees - of subsistence, on land someone else owned. Every reserve has a warden, and the warden serves whoever holds the title deed. Gates does not say who either will be.

Policy holds four levers here - an accelerator, a brake, equity and a tax - and each can be pulled on the machine or on the human. It is worth laying all eight cells out, because the full map shows how little of it Gates explores.

Notice where his essay lands: two cells claimed, two brushed in passing, and the equity row - the only one that changes who owns anything - never entered. He is eloquent about who will lose; but a safety net compensates the losers after the fact, while ownership decides the outcome before it. Six thousand words, conducted in the defensive half of the space.

Now read the columns, lever by lever. The machine’s side is open road. The accelerator is floored: the tax code rewards every firm that buys software instead of hiring. There is no brake: nothing restrains where the machine may work. Equity sits with a small circle of owners. And the tax never comes: the machine’s output goes untaxed.

The human column is the mirror image. The only accelerator is whatever the levy might one day fund. Reserved occupations put a brake on human adaptation. Equity extends no further than one’s own labour. And the tax lands on every wage, automatically.

Gates is right that the tax code nudges every employer toward the machine. By the European Commission’s own data, big tech was taxed more lightly than the shops it was replacing during the last platform shift. The tax incentive thereby accelerated the disruption. But his signature proposal is a brake, and in an incumbent’s hand a brake is a moat: compliance thresholds do to challenger models what licensing does to challenger workers. That is the cell the machine’s owners concede most gladly. If the Human Reserve passes, it will pass for the same reason it fails: it costs the machine nothing.

The brake and the tax defend: they restrain the machine and change no ownership. The accelerator and equity build: the accelerator decides what gets made, equity decides who owns it. The warden works with the building levers.

Gates only focuses his attention on the defensive levers. The accelerator is the true opposite of a reserve, and he touches it only as spending. Equity he does not reach for. Yet equity is the whole predicament: the machine is owned by a handful, and for most, the only equity they have is their labour, which is being repriced by AI’s competition.

The ad-hoc policy improvisations have begun. Washington converted its chip subsidies into a 10% Intel shareholding last August, and now seeds each American newborn with a thousand dollars in an index fund. Crude instruments, but proof the lever exists and pulls. A population with shares in the abundance needs no fence around its work. How the shares reach everyone else - sovereign stakes, compute dividends, claims that survive dilution - is a separate topic.

For the boardroom. The useful preparation is a map of your own task mix: which cells of the grid you already occupy, which your customers and insurers are about to choose for you, and how fast the people you keep are moving from reserved work to “augmented work”. That is a capital-allocation exercise, and it fits on one page.

For investors. A reserve is a rent while it stands and a market handed to outsiders when it falls. Both legs are underwriteable - scarcity pricing inside the fence, the unencumbered challenger outside it - and the discipline is knowing which year of the fireman’s thirteen you are in. The same grid that exposes Gates’s essay prices a portfolio.

Lion Strategy advises boards and investors on both. To discuss what this means for your task mix or your portfolio: contact@lionstrategy.com.

Part of the Intelligence Economy series, with Judgement. Exposed. and The AI rung gap.

To map your own task mix across the grid - or price both sides of the fence for a portfolio - start a conversation.

Talk it through with Elliot ←︎ BACK TO THE SERIES