Bill Gates' "Human Reserved" Jobs: Will AI Take Yours?

On August 26, 2026, Bill Gates published a 6,000-word essay with a title that reads like a warning label: The turbulent AI era is here. The choices we make now are critical. The headline everyone ran was "Gates wants up to 40% of jobs reserved for humans." The part almost nobody explained is what he actually means by that — and it is close to the opposite of every "jobs AI can’t replace" list you have ever scrolled. Gates is not describing jobs AI can’t do. He is describing jobs AI can do, that we decide to keep for people anyway — the way a nature reserve is land you could build on and choose not to. That flips the question most people are typing into a search box. "Will AI take my job?" stops being a technical prediction about what software can manage and becomes a political one about what a society is willing to let go. Below: what the essay says in plain words, the jobs he names on both sides of the line (his "vanish first" list includes the exact careers people retrained into), the tax idea underneath it, the honest holes in the plan, and a five-question self-test that beats any "will AI take my job calculator" — because the calculators are guessing, and this page will show you why.

Jake sent me the headline at 7 a.m. with one line: "So is the coding course a waste?" His son starts a two-year software program this fall. Jake runs a repair shop; he has spent eighteen months telling the kid that fixing phones is a dead end and that writing software is the safe bet, and here was the co-founder of Microsoft — the man who is software, as far as Jake’s generation is concerned — putting software engineering on the list of jobs that disappear first. He wanted a yes or a no. What he got, over two coffees, is this page. The short version I gave him: the course is not a waste, but the job it was supposed to lead to has changed shape, and the shape it changed into is the thing to aim at. The long version is below, and it is not just for Jake’s son. It is for the 55-year-old in construction Gates mentions by name, the paralegal who just paid off her certificate, and the call-center lead whose team has been shrinking every quarter without anyone saying the word out loud.

Ethan: "Everybody read ‘Human Reserved’ as a list of safe jobs, and it isn’t. Think of a national park. Nobody protects Yellowstone because you can’t build a mall there. You obviously can — that’s the point. It’s protected because someone decided a mall wasn’t worth what you’d lose. Gates is saying the same thing about a nurse telling you bad news, or a person watching your kid: a machine could do it, probably cheaper, and we should draw a fence around it anyway. That’s not a prediction about technology. It’s a decision about what kind of place you want to live in. And decisions, unlike predictions, are things you can actually argue with."

What Gates actually wrote (not what the headlines said)

The essay went up on Gates’s own site on August 26. It is long, it is unusually blunt for him, and it is built on one claim that everything else hangs from: this transition is different from the last ones because, in his words, "the technology can substitute for human cognition" — not muscle, not routine, but thinking. Earlier automation waves took a few generations to work through the economy. He expects this one to play out "over the course of a decade rather than a few generations," hitting "workers in industries such as law, customer service, medicine, software and manufacturing." And he is explicit that new jobs will not automatically fill the hole: "There will be some new jobs, but without the right policies there will be far fewer than exist today."

Then the line that made the essay news rather than commentary: "Unfortunately, right now we are not preparing for it. I don’t see evidence that leaders, experts and communities are confronting the challenges adequately. There is no plan to ease the entry into the AI era." He frames the stakes as binary — "AI will either be the greatest equalizer ever invented, or the worst source of injustice" — and calls answering the question "the world’s top priority." He also does something rare for a technology billionaire writing about technology: he tells you where he stands financially. He says he has benefited enormously from the industry, still has financial ties to it, is working with Microsoft and other AI companies, and that readers "will have to decide for themselves whether this clouds my view." Keep that disclosure in mind; we will come back to it, because it cuts both ways.

The essay names three risks and offers three responses. The risks: work disappearing, especially entry- and mid-level work and especially for young people; criminals and hostile states gaining capabilities they never had; and children’s development being dulled by AI companions that never push back. The responses: new governance bodies at home and internationally, a set of jobs deliberately kept for humans, and a change to how AI and robots are taxed. The rest of this page takes those apart one at a time, in the order a worried person actually needs them — jobs first.

"Human Reserved" — the idea most coverage got backwards

Here is the definition in Gates’s own framing: Human Reserved is work that machines will be fully capable of doing, but that we decide to keep for people anyway, and the model he reaches for is a nature reserve — land where we could build roads and buildings, and choose not to. Read that twice, because it is the single most misreported sentence of the week. Every "jobs AI can’t replace in 2026" article — and there are dozens, listing plumbers, surgeons, therapists, electricians, clergy — is making a capability argument: AI can’t do this, so you’re safe. Gates is making a choice argument: AI will be able to do this, and we should fence it off regardless. Those are not the same list, and the difference matters enormously to the person deciding what to train for.

He gives two reasons a job might end up inside the fence. The first is moral. His example: "imagine a robot giving you the awful news that you have an incurable disease. There’s no technical reason why it couldn’t. Yet it shouldn’t." Childcare and jury service, he told reporters, clearly fit — there is something about a jury of your peers, or a person raising a child, that is the point of the job, not a limitation of it. The second reason is economic, and it is the one that applies to most of the people reading this: "We might set something aside as Human Reserved for economic reasons… we may do it because allowing machines to take over a certain role will displace a large number of people who can’t easily change jobs." His illustration is a 55-year-old who has worked in construction their whole career: you cannot tell that person to go work at an elder-care facility "and expect them to find it fulfilling." Some reservations, in other words, would be permanent; others would be a buffer — a role held for humans for years or decades while a generation that cannot retrain ages out of it. He writes that the Human Reserved domain "will evolve over time" and that we should consider "setting aside some jobs now and phasing in AI slowly over years or decades."

And the 40%? That number is the ceiling, not the plan. Asked how far the idea could stretch, he put the most aggressive version at roughly 40% of jobs and described that figure as about "as high as I can get." So the honest headline is not "Gates wants 40% of jobs reserved." It is "Gates thinks a society could, at most, choose to protect four in ten jobs, and hasn’t decided which." He is candid that he has not decided either: "Who gets to decide what we reserve for humans? What criteria should we use?" are questions he raises and leaves open. We will get to why that is a real weakness. First, the lists.

The jobs he names — on both sides of the fence

Look at the first row again, because it is the one that stopped Jake cold. Customer support, sales, software engineering, paralegal. Three of those four are the careers a decade of advice told people to move toward. "Learn to code" was the retraining slogan of the 2010s. Paralegal certificates were sold as the affordable door into law. And customer support was the job that absorbed everyone the last wave of automation pushed out of factories and back offices. The uncomfortable pattern is that the safe harbor of the last transition is the exposed coastline of this one, and Gates is the first person of his stature to say so without hedging. That is also why a "careers AI can’t replace" list from 2019 is worse than useless now: the list was built on the muscle-versus-mind rule, and this wave broke the rule.

Why software engineering is on the "first to go" list — and what that really means

This deserves its own section because it is the one people will argue with hardest, and because the argument is usually about the wrong thing. Nobody serious — not Gates, not the labs — is claiming the profession of building software vanishes. What is vanishing is a specific rung: the entry-level and mid-level work that a new graduate used to be hired to do while learning the rest. Gates puts it plainly: "The jobs at most risk are entry- and mid-level, and the new jobs being created will mostly require skills that take many years to learn." That is the trap. The ladder has lost its bottom rungs, and the top rungs are still there.

He backs it with data rather than vibes, and it is worth knowing which data, because you will see it misquoted. The study is from the Stanford Digital Economy Lab (Brynjolfsson, Chandar, and Chen, November 2025), built on actual payroll records from ADP — monthly, individual-level, millions of workers — not surveys of what managers say they plan to do. Its headline finding: early-career workers aged 22 to 25 in the occupations most exposed to AI saw a 16% relative decline in employment. Older workers in the same occupations did not. And the detail that matters most to you: employment fell where AI automates tasks and grew where AI augments work. That split — automate versus augment — is the most useful two-word test in this whole debate, and we build the self-test around it below.

So what did I tell Jake about the course? Three things. One: a program that teaches someone to produce code that an assistant now produces in seconds is training for the rung that is gone. Two: a program that teaches someone to specify, review, debug, integrate, secure, and take responsibility for code — the parts that require knowing what "correct" means for a real business — is training for the rungs that remain, and those pay better than the old bottom rung ever did. Three: the kid should spend the two years building things that run, in public, with his name on them, using every AI tool available, because the person who gets hired in 2028 is the one who can show they direct the machine rather than compete with it. If you want the mechanics of what these tools actually are and are not, our explainers on what a large language model really does and what an AI agent actually is are the plain-English versions; the second one includes the compounding-error math that explains why the machine still needs a human holding the leash.

The token tax and the robot tax, with the numbers underneath

The second proposal is a tax, and Gates is not shy about its size: "I’m talking about a change to the tax system that’s greater than any in my lifetime." The logic starts from something most people have never noticed about how businesses are taxed. In his words: "Right now, if you’re an employer and you hire someone, you pay payroll taxes on their earnings. But if you buy a robot, you can usually write it off right away as a business expense." A human costs the employer their wage plus a payroll-tax layer on top; a machine that does the same work is a deductible purchase. The system, he says, "nudges you toward replacing people with machines" — not out of malice, just arithmetic.

Two levers, then. The robot tax ends the immediate write-off and treats a robot bought to replace labor more like the labor it replaces. The token tax is the new part: a levy per token on output a model produces — the unit that cloud AI is already metered in — so that an employer replacing a team with an agent pays something toward the people it displaced. The stated purpose is not to stop automation; he is clear that he does not think it can be stopped. It is to "slow the rush away from human labor a little and raise money for retraining and a stronger safety net." He also owns the history: "I proposed a robot tax years ago and most of the reaction was that it was a strange idea." That was 2017. The reaction this time has been quieter, which tells you something about the last nine years.

That token-tax objection in the middle row is not a hypothetical; it is already the shape of the market. This week alone, IBM shipped a family of open reasoning models under a license that lets any company run them on its own servers for nothing per token, forever. Tax the cloud token and you have just written a subsidy for the local one. Gates does not address this in the essay, and it is the first thing engineers raised when it was published. A token tax that works would have to be a tax on compute or on displacement, not on the API bill — and both of those are much harder to write into law.

The three risks he says nobody is planning for

Risk one: work. Covered above, but with one addition that makes it concrete rather than abstract. Since 2023, U.S. employers have cited AI as a reason in roughly 184,000 announced job cuts, per the outplacement tallies the press has been quoting alongside the essay; in July 2026 alone, close to 11,000 cuts named AI, about a third of that month’s total. Call-center employment in the U.S. is running well below its long-run trend. None of that is a forecast. It is the rearview mirror. Gates’s claim is that the windshield looks worse, and that the people being hit first are the ones with the least cushion — the young, the entry-level, and the "accounting worker who’s replaced by a bot." We watched a small version of this happen in real time yesterday: Amazon is shutting down Mechanical Turk, the platform where half a million people did the human micro-tasks that trained the machines now making those tasks unnecessary. The workers who taught the model are its first alumni.

Risk two: misuse. "The smartest cybersecurity experts I know are scared about the next few years," he writes, and the sentence that follows is the whole problem in one line: "the same AI model that can find a flaw in software so a company can fix it can also help a criminal exploit it." He escalates from there to autonomous weapons and loss-of-control scenarios, but the near-term version is the one that reaches ordinary people: phishing that reads like your manager, voice clones that sound like your daughter, and fraud at a scale and polish that used to require a team. If you read our piece on how scam pages hide on legitimate infrastructure, you already know the con does not need new technology to work — it needs volume and believability, and both just got cheap.

Risk three: children. This is the section parents should read twice. Gates points at AI companions — apps built to be endlessly agreeable — and at early research linking heavy use to isolation, plus a preliminary survey suggesting heavier AI use went along with less critical thinking. He is careful to call the evidence early. But the design concern does not need a final study: a companion that never disagrees, never gets bored, and never makes a child work for an answer is training the opposite of resilience. His proposed fix is the most practical paragraph in the essay, and we quote it in the students section below, because it is a rule you can apply tonight.

"Will AI take my job?" — why every calculator is guessing

Type "will AI take my job" into a search box and the autocomplete offers you a calculator, a website, a documentary, and a meme, in roughly that order. The calculators all descend from the same 2013 academic exercise that scored occupations by how "routine" their tasks looked, and they have three problems. They score the job title, not your job — two people with the same title can have opposite exposure depending on whether the AI in their office does their tasks or drafts them. They were built before models could write, code, and reason, so their idea of "routine" is a decade stale. And, the big one: they measure capability, and this page has spent two thousand words showing that capability is only half the answer — the other half is what employers, regulators, and voters decide to allow. No calculator has a field for that. 

What actually predicts whether you get replaced is the automate-versus-augment split from the Stanford payroll data. So instead of a calculator, here is a five-question test built on it. Answer honestly; the pattern of your answers is the reading.

1. When you use AI at work, does it do the task or draft the task? If a tool produces the finished thing and you send it, that is automation. If it produces a starting point that you materially change, that is augmentation. Augmented roles grew in the data; automated ones shrank.
2. Who is accountable when your output is wrong? If the answer is "me, by name, with consequences," the job has a responsibility layer that companies are slow to hand to a machine — that is Gates’s whole second-wave argument about loans, diagnosis, and triage. If nobody is accountable and errors just get fixed downstream, the layer is thin.
3. Is your output measured in units the machine is priced in? Tickets closed, words written, lines committed, calls handled per hour — if your performance review counts things a model produces by the thousand, you are in the first wave. If it counts outcomes that need a person in a room, you are not.
4. Could your employer replace your team with one senior person plus tools? That is the actual layoff shape of 2025–2026: not "the AI took the job," but "one experienced person now does the work of five juniors." If you are the one senior person, your value went up. If you are one of the five, it did not.
5. Is your job on the moral side of the fence? Does the person you serve want a human there — a child, a patient, a jury, a grieving family? That is the Human Reserved test, and it is the only one of the five that depends on society rather than your boss. It is also the one where you can do something: those jobs stay human only if people insist.

Three or more "automated" answers means the wave is already at your door and the move is to change what you do inside the same job, this year, not next. Mostly "augmented" means you have time — use it to become the senior person in question four. And if question five is a strong yes, your job’s future is a political question, and the honest advice is to be part of answering it rather than waiting to be told.

Job by job: risk, the real reason, and the move

The table below is deliberately not a "safe jobs" list. It is the job families people ask about most, with Gates’s placement where he gave one, the honest reason behind the risk, and the single most useful move for someone in that seat. The "why" column matters more than the risk column — it is what tells you whether the reason is capability (which will only grow) or choice (which can be fought).

For developers and IT people specifically

Most readers of this site are on the technical side, so here is the un-softened version. The essay is right that the junior software rung is going, and it is right for a reason that is under your control: the work that vanished is the work that was already defined by someone else. A ticket that says "add a field, write the migration, update the tests" is a fully specified task, and fully specified tasks are what the tools eat. The work that is growing is the work of turning a vague, risky, expensive business problem into a set of fully specified tasks — and then being the person whose name is on the result when it breaks at 3 a.m. That has always been the senior job. What changed is that the apprenticeship that used to lead there has been shortened from years of ticket work to however long it takes you to build real things with the tools and take responsibility for them.

Three concrete moves. First, own a surface nobody can automate the accountability for: security, infrastructure cost, data integrity, compliance. These are the areas where "the AI did it" is not an acceptable sentence in a post-mortem, which is exactly why they hold value. Our free AWS series exists partly because cloud infrastructure is one of the few technical ladders that still has its bottom rungs — you can build a real, accountable system for $0 and put it on a resume. Second, become the reviewer. Every team that replaced five juniors with one senior plus tools is now desperate for people who can read machine-written code for the mistakes the machine makes confidently. Third, and this is the one people skip: learn to explain technical risk to non-technical people in plain sentences. The second-wave jobs — the ones regulation is protecting — are protected precisely because someone has to stand in a room and answer for a decision. Be that someone.

For students and parents: the "productive struggle" rule

Buried two-thirds of the way into the essay is the most immediately useful thing in it, and it is about homework. Gates describes the AI tool a student should have: one that preserves what researchers call "productive struggle" — the cognitive work that builds understanding. In his words: when a student first meets a new idea, the AI gives substantive explanations and offers both questions and answers; later, when it is checking their comprehension, it "holds the answer back and helps them arrive at it on their own." That is a design spec for a tutor. It is also, flipped around, a rule for a kid with a chatbot tonight: use it to explain, never to finish. Explaining a concept builds the muscle. Finishing the assignment deletes the reason the assignment existed.

For Jake’s son and every student choosing a field, the essay implies a rule of thumb that no course catalog will print: pick the version of your field where you are accountable for outcomes, not for output. A nursing program over a medical-coding certificate. A trade license over a data-entry track. A computer-science degree that ends in you owning systems, over a bootcamp that ends in you producing tickets. And, whatever the field, graduate with proof you can direct the machine — a portfolio of things that actually work, built with every tool you could get your hands on. The Stanford data says the 22-to-25 cohort in exposed jobs is down 16%. It does not say the 22-to-25 cohort who can be the senior person in question four is down at all. Nobody has that data yet, because those people are only now graduating. Be one.

What to do this year, whatever your job

1. Run the five-question test on your actual week, not your job title. Write down which of your recurring tasks a tool could produce end-to-end today. That list is your exposure; the rest of your week is your value.
2. Move one task from "automated" to "augmented" every quarter. If a tool can draft it, become the person who specifies it, reviews it, and is accountable for it. Tell your manager you are doing this; it changes how you are counted.
3. Build a six-month cushion before you need it. Gates’s whole argument is that this transition is faster than the last ones. A layoff in a first-wave job is now a when-not-if planning item, and cash buys you the time to retrain toward a rung that exists.
4. Learn the tools better than the people who fear them. The 2026 layoff shape is one senior person plus tools replacing a team. The person who kept their job was usually the one who was already using the tools to do the work of three.
5. If you are in a Human Reserved job, defend the fence. Nurses, teachers, carers, and everyone who serves a person face-to-face: your protection is a social decision, not a technical one. Unions, licensing boards, professional bodies, and ballots are where that decision gets made. Gates has handed you the argument; the job is to use it.
6. Keep one eye on the tax debate. If a token or robot tax funds retraining and a safety net, the retraining money is yours to claim. If it does not pass, the cushion in step three is the only safety net there is.

The honest holes in the plan

Balance, because this site does not do fan mail. Gates asks the two hardest questions about Human Reserved himself — who decides, and by what criteria — and then leaves them on the table. That is not a small gap; it is the whole mechanism. A reserve with no boundary is a slogan. Enforcement is the second hole: a rule that a role must stay human is only as strong as the audit that checks it, and a firm that quietly automates 80% of a "reserved" job while keeping one person as the nameplate has complied with the letter and gutted the point. The third hole is the one economists raised within hours: a country that reserves 40% of its jobs is competing with countries that reserve none, and Gates’s answer — that the governance has to be international, in the mold of nuclear inspections, aviation rules, and the ozone treaty — requires a level of U.S.–China cooperation that does not currently exist on anything.

The token tax has the loophole we covered: local models produce untaxed tokens, and the open-weight market is exploding precisely to give companies that option. The robot tax has a definitional problem — is a forklift with a camera a robot? — that his 2017 version never solved either. And then there is the disclosure. He tells you he still has financial ties to the AI industry and is working with Microsoft and others, and asks you to judge for yourself. Judge both ways: it means his "the jobs are going" claim comes from someone with a front-row seat and no incentive to exaggerate the downside of his own industry; it also means a plan that taxes and fences AI while leaving the labs themselves standing is a plan the labs can live with. Both readings are fair. What is not fair is the lazy one — that this is a billionaire being dramatic. The Stanford payroll numbers are not his. The layoff counts are not his. The essay’s value is that it puts a name that people trust on data they were already ignoring.

"What happens when AI replaces all jobs?" — the question under the question

It is one of the most-searched forms of this whole topic, and the essay’s honest answer is that nobody knows, including him — and that this is exactly why the choices being made now are the ones that matter. Every previous transition created new categories of work to absorb the displaced; the unresolved argument among serious people right now is whether a technology that substitutes for thinking leaves any "new space" for humans to move into, or whether this time the space simply does not appear. Gates plants himself on the pessimistic side of that argument for planning purposes — "I’d love to be convinced that I’m wrong about the job market, but I’m not" — while insisting the outcome is a choice rather than a fate. Human Reserved is what a society looks like if it decides some work stays human on purpose. The taxes are how that society pays for the people the machines displaced. And the international bodies are his admission that no single country can hold the line alone.

The most useful way to hold all of this, for a person with a job and a mortgage rather than a foundation and a platform, is the one Ethan gave up top. Predictions you can only wait for. Decisions you can argue with. Gates has reframed the biggest economic question of the decade from the first kind into the second. That is either the most hopeful thing in the essay or the most frightening, depending on how much you trust the people who will be making the decisions — which is, in the end, the reason he wrote 6,000 words to the rest of us instead of a memo to them.

FAQ — Gates, Human Reserved, and your job, answered straight

What does Bill Gates mean by "Human Reserved" jobs?

Work that machines will be fully capable of doing but that society decides to keep for people anyway — for moral reasons (childcare, jury duty, delivering a terminal diagnosis) or economic ones (roles held by large numbers of people who cannot easily retrain). His analogy is a nature reserve: land you could build on and choose not to.

Did Gates really say 40% of jobs should be reserved for humans?

He gave roughly 40% as the ceiling of the most aggressive version and called it about as high as the idea could go — not as a target. He has not said which jobs, and openly asks who should decide and by what criteria.

Which jobs does Gates say AI will replace first?

Customer support, sales, software engineering, and paralegal work in the first wave; loan assessment, data analysis, and patient triage in a second wave; and construction and hospitality shifting toward automation by the end of the decade as dexterous robots improve.

Will AI take my job?

It depends less on your title than on whether AI automates your tasks (produces the finished output) or augments them (drafts something you materially change and are accountable for). Payroll data shows employment fell in the first case and grew in the second. The five-question self-test above is built on that split.

Is there a reliable "will AI take my job" calculator?

No. The calculators score job titles using a routine-task model from 2013, built before AI could write or code, and none of them account for the political and regulatory side of the question that Gates’s essay is about. Use them for curiosity, not for decisions.

What jobs can AI not replace?

Honestly: fewer than the lists claim, and the number shrinks every year. The durable categories are licensed physical trades in irregular environments (electricians, plumbers, HVAC), roles where the person served insists on a human (care, teaching, therapy, juries), and roles where a named human is legally accountable for the outcome. The first is a capability wall; the other two are choices society has to keep making.

Are there high-paying jobs AI can’t replace?

The well-paid durable roles share a pattern: accountability plus judgment plus a person who wants a human. Senior engineers who own systems, security and infrastructure leads, specialist trades, surgeons and senior clinicians, and complex-deal sales. The pay is for the responsibility, which is exactly the part firms are slowest to hand to a machine.

Why does Gates put software engineering on the "vanish first" list?

Because the entry- and mid-level rung — fully specified ticket work — is what current tools do best, and payroll data shows a 16% relative employment decline for 22-to-25-year-olds in exposed occupations. The senior work of specifying, reviewing, securing, and being accountable for systems is growing, not shrinking.

What is the "token tax" Gates proposed?

A per-token charge on output a model produces from metered services, paired with ending the instant tax write-off for robots bought to replace workers. The goal is to slow the rush away from human labor slightly and fund retraining and a safety net. The biggest objection is that models run on a company’s own hardware produce untaxed tokens.

Hasn’t Gates proposed a robot tax before?

Yes, in 2017, and he notes that most of the reaction then was that it was a strange idea. The new version adds the token tax and frames the combined change as larger than any tax reform in his lifetime.

What are the three risks in the essay?

Work disappearing, especially entry- and mid-level jobs and especially for young people; criminals and hostile actors gaining capabilities they never had, up to autonomous weapons and loss-of-control scenarios; and children’s development being dulled by AI companions that never push back.

How fast does Gates think this happens?

Over roughly a decade, versus the few generations earlier automation waves took. First-wave cognitive jobs are already moving; physical work shifts toward the end of the decade.

Will my job be replaced by AI if I work in IT?

Routine ticket and helpdesk work shrinks; ownership of security, infrastructure, cost, and data integrity grows, because "the AI did it" is not an acceptable line in a post-mortem. Gates’s own cybersecurity warning makes defenders more needed, not less.

What should students study if AI is taking entry-level jobs?

The version of any field where you are accountable for outcomes rather than output, and a path that ends with proof you can direct AI tools to build things that work. Use AI to explain, never to finish — Gates’s "productive struggle" rule.

What is the strongest criticism of Human Reserved?

Nobody — including Gates — has said who decides which jobs qualify, how it would be enforced, or how a country that reserves jobs competes with one that does not. His answer is international governance, which currently does not exist for AI.

Does Gates have a conflict of interest here?

He says so himself: he has benefited enormously from the industry, still has financial ties, and is working with Microsoft and other AI companies, and he asks readers to judge whether that clouds his view. The data he cites (Stanford payroll records, layoff tallies) is not his own.

What happens when AI replaces all jobs?

Nobody knows, and the essay’s point is that the answer depends on decisions made now: whether some work is reserved for humans, whether automation is taxed to fund a safety net, and whether countries coordinate. Gates plans for the pessimistic case while insisting the outcome is a choice.

Where can I read the essay?

On Gates’s own site, published August 26, 2026, under the title "The turbulent AI era is here. The choices we make now are critical." It runs about 6,000 words; budget half an hour.

Revision note. Written August 27, 2026, the day after the essay appeared, from the essay itself and the interviews Gates gave alongside it; the employment figures come from the Stanford Digital Economy Lab’s payroll study and the layoff tallies quoted in that coverage, and the job placements in the tables are his where he gave them and marked as ours where he did not. If he revises the proposal — he says the Human Reserved list should evolve — this page will follow. And to Jake, and to everyone who read the headline at 7 a.m. and felt the floor move: the course is not a waste, the trade is not a dead end, and the job you have is not gone. It is changing shape. The people who come through this are not the ones who guessed the safe job; they are the ones who noticed what they are accountable for and leaned into it. You have already done the noticing. That is the hard part.