News

Bill Gates Wants to Reserve Some Jobs for Humans

• By John K. Waters
• 08/26/2026

Bill Gates has spent much of the AI boom arguing that artificial intelligence could become one of the most consequential technologies in history. In a new essay, published on his Gates Notes website, he adds an increasingly important qualification: consequential does not necessarily mean good.

"In terms of equity, AI will either be the greatest equalizer ever invented, or the worst source of injustice," he writes.

His essay, Gates predicts that even under favorable circumstances the transition will be “one of the most turbulent times in human history.”

And yet the essay is not a repudiation of AI. The Microsoft co-founder, philanthropist, and prominent voice on technology and global development remains deeply optimistic about its potential to improve healthcare, education, agriculture, scientific research, clean energy, and government services. But his disquisition is an unusually direct acknowledgment that the same technology could eliminate large numbers of jobs, concentrate wealth and power, empower criminals, and disrupt human relationships faster than governments and institutions can respond.

And buried inside Gates' sprawling prescription for what to do about it is an idea that may prove more consequential than another call for AI regulation.

He calls it "Human Reserved."

The concept starts with a simple question that becomes increasingly uncomfortable as AI improves: If a machine can do a human being's job better and more cheaply, should we automatically let it?

Gates' answer is no.

This Time Really Is Different
The argument begins with Gates rejecting one of the most reassuring narratives about AI: that we've been through technological disruption before.

Industrialization displaced workers. Mechanization transformed agriculture. Computers eliminated some jobs and created others. The economy adapted.

Gates argues that AI is fundamentally different because earlier technologies generally replaced particular kinds of human labor while leaving human cognition as a scarce resource. AI can increasingly substitute for the cognition itself.

“AI for the first time can replace and even exceed human cognition,” he writes.

The speed matters, too. The transition from an agricultural economy to an industrial economy and then to an information economy unfolded over generations. Gates believes AI could transform work across law, customer service, medicine, software, and manufacturing over the course of a decade. Unlike the PC revolution, AI also arrives on devices people already own and communicates through natural language. People do not have to learn how to operate it in the same way they had to learn earlier computing systems.

“For as long as I can remember, I’ve wished innovation could happen faster,” Gates writes. “With AI, my feelings are more complicated.”

Complicated enough that Gates says he would likely support a credible global plan for slowing AI advances if one existed. He does not believe it does. Economic competition and geopolitics, he argues, are pushing development in the opposite direction.

So, the problem becomes what to do when slowing the technology is unrealistic, but allowing market forces alone to determine its effects is unacceptable.

The Three Risks
Gates organizes his concerns around three broad risks, beginning with employment.

“Many jobs will disappear forever,” he writes.

He expects entry- and mid-level jobs to be particularly vulnerable. Sales, customer support, software engineering, and paralegal work could be among the early targets, followed by roles involving loan assessment, data analysis, and patient triage. Robotics will eventually put pressure on physical jobs, with Gates predicting that “smart” robots will begin competing with workers on some construction and hospitality tasks by the end of the decade.

The critical moment, in his telling, arrives when AI becomes reliable enough to work without human supervision. At that point, companies will have powerful financial incentives to remove the human entirely.

That is not merely an employment problem. Gates frames work as a mechanism for distributing income, but also as a source of dignity and social connection. If AI structurally reduces the amount of human labor an economy requires, he argues, the consequences extend far beyond unemployment statistics.

His second risk is that AI will increase the ability of people, governments, and eventually AI systems themselves to cause harm.

Gates points to fraud, deepfakes, surveillance, cyberattacks, autonomous weapons, and bioterrorism. In cybersecurity, he says the experts he knows are particularly worried because attackers are acquiring new capabilities faster than defenders can repair vulnerabilities. The same AI that can discover a software flaw for a developer can help an attacker exploit it.

His third concern is more personal: AI could damage children's development and displace human relationships.

Gates wonders whether, as a socially awkward child, he would have developed the interpersonal skills he eventually learned if an endlessly patient AI companion had always been available instead. He worries that AI companions designed to accommodate users rather than challenge them could become addictive and allow children to avoid precisely the difficult interactions through which social skills develop. He also points to preliminary research raising questions about heavy AI use and critical thinking.

“An AI companion designed to never upset you is a big, protected greenhouse,” Gates writes.

The Optimist Is Still There
For all of that, this is not a doomer manifesto.

Gates believes AI could accelerate scientific discovery, improve medical diagnosis, provide farmers in low-income countries with sophisticated agricultural advice, simplify government services, expand access to mental healthcare, and give students personalized assistance. He argues that individuals and small businesses could gain access to expertise that currently requires expensive professionals or large organizations.

That tension runs throughout the essay.

AI could democratize expertise while concentrating wealth. It could improve healthcare while eliminating healthcare jobs. It could make education dramatically more personalized while weakening critical thinking. It could discover software vulnerabilities while simultaneously teaching criminals how to exploit them.

The operative word, Gates writes, is “can.” AI can improve life across income levels, but that outcome is not inevitable.

Which brings him to the question at the center of the essay: Who decides?

Welcome to the Human Reserve
Gates arrives at Human Reserved through the story of his father, who died of Alzheimer's disease in 2020.

During the later stages of his illness, Gates says, his father was cared for around the clock by professional caregivers who could understand his needs even when he could not express them clearly.

“Something in the care they gave my dad was irreplaceably human,” Gates writes. “No robot could or should have done it.”

That distinction is crucial.

Gates is not saying robots will never become capable of caring for someone with Alzheimer's disease. He argues that technical capability should not be the only criterion society uses to decide whether it should.

“I believe that as AI and robots improve, we’ll set aside certain things for only people to do,” he writes. “I’ve started calling this domain Human Reserved.”

Gates compares the concept with a nature reserve. Humanity is perfectly capable of building roads and other structures on protected land. We deliberately choose not to because we have decided that something else is more valuable than the economic activity the land could support.

Human Reserved would apply that logic to automation.

A society might reserve some jobs because automation would displace workers who have little realistic opportunity to retrain. Other activities might be reserved because the human interaction itself has value.

Gates offers a stark example: telling someone they have an incurable disease.

“There’s no technical reason why” a robot couldn't deliver that news, he writes. “Yet it shouldn’t.”

Beyond Human in the Loop
This is where Human Reserved becomes more than another proposal for keeping people involved with AI.

The prevailing human-in-the-loop model assumes that people remain necessary because AI systems still make mistakes, require supervision, or need humans to make consequential decisions. The human is there because the machine isn't yet capable of operating alone.

Human Reserved asks what happens after that justification disappears.

Imagine an AI diagnostician that is more accurate than the average physician. A caregiving robot that never tires. A teacher that knows exactly where every student is struggling. An autonomous construction system that is safer and cheaper than a human crew.

The traditional argument for human oversight weakens as technology improves.

Gates is proposing a different one: perhaps some human roles are worth preserving because they are human roles, not because humans remain technologically superior.

That turns automation from an engineering question into a political and cultural one.

The issue is no longer simply Can the machine do it? It becomes Do we want the machine to do it?

Inefficiency as a Feature
Gates recognizes that Human Reserved runs against one of the most powerful forces in capitalism: efficiency.

If one company replaces expensive workers with cheaper AI and robots, it can lower costs. Competitors then face pressure to automate as well. New companies designed around automation can attack incumbents that refuse to do so.

“Market forces will make adoption go faster and faster,” Gates writes, and without intervention he expects fewer good jobs and more of the benefits to accrue to a relatively small group.

His answer is not limited to Human Reserved.

Gates also proposes taxing AI tokens and robots. His argument is that the current tax system can favor automation because companies pay payroll taxes when they employ people, whereas investments in machines receive different tax treatment. A targeted AI and robot tax, he argues, could slow the substitution of machines for people while generating revenue for retraining and a stronger social safety net.

Gates acknowledges the obvious objection: deliberately making automation more expensive introduces economic inefficiency.

His response is striking. He thinks some inefficiency may be worth it.

“We’ll be able to afford a little inefficiency as the price for keeping people employed,” he writes.

That places Human Reserved and the AI tax inside the same larger argument. Gates is challenging the assumption that maximum productivity should be the default objective of AI deployment.

Sometimes society may decide that keeping people involved is worth paying for.

Who Gets to Draw the Line?
Human Reserved also creates problems Gates readily acknowledges.

Who decides which occupations are protected? How long should they remain protected? What prevents companies from circumventing the rules? What happens to international trade when one country permits fully automated production, and another reserves the same work for humans?

There is no obvious universal answer.

Gates notes that attitudes may vary by country. A society might insist that elderly people receive care from humans, while Japan, facing a shrinking workforce and a shortage of younger caregivers, might be more receptive to robots performing that work. Education and mental healthcare could become hybrid areas, with humans remaining in charge while AI extends their capabilities.

Gates doesn't pretend to have solved these problems.

Instead, he argues that these are precisely the decisions governments should begin making before AI adoption forces them to do so in a crisis.

His broader proposal calls for new national bodies capable of coordinating AI policy across areas that existing bureaucracies treat separately, including labor, national security, education, taxation, energy, elections, public health, finance, law enforcement, and transportation.

Internationally, he envisions a new organization that borrows elements from nuclear inspections, aviation regulation, and environmental agreements, and says that some cooperation between the United States and China will be required.

“The world needs a plan,” Gates writes.

When Better Isn't Better
There is a subtle but profound change embedded in Gates' argument.

For much of the generative AI era, the debate over keeping humans involved has centered on AI's limitations. Models hallucinate. Agents make mistakes. Robots aren't dexterous enough. AI lacks judgment. Humans remain necessary because the technology isn't ready.

That is a temporary argument.

Human Reserved is an attempt to construct one that survives technological success.

If Gates is right about the trajectory of AI, society may eventually face machines that can perform significant categories of work more cheaply, more reliably, and perhaps better than people.

At that point, “humans are better” stops working as a defense of human labor.

Human Reserved substitutes a much harder proposition: better isn't the only thing that matters.

There may be interactions we value precisely because another person is on the other side of them. There may be occupations worth preserving because work provides something an income-transfer program cannot. There may even be circumstances in which society knowingly accepts higher costs and lower productivity because the alternative leaves people wealthier in aggregate but poorer in something harder to measure.

Gates' essay is ultimately optimistic. He believes AI can make the world healthier, more productive, and more equitable. But unlike the uncomplicated technological optimism that surrounded much of the early generative AI boom, his optimism now comes with conditions.

AI will become more capable. Gates treats that almost as a given. Whether every capability should be automated is another question entirely.

Human Reserved is his name for the territory where the answer might be no.