AI and Jobs by 2040: Why Mass Unemployment Isn't Coming — and What's Coming Instead Author: Дністер Published: 2026-08-15T03:01:18.000Z Language: en URL: https://neurodrift.org/en/blog/robota-do-2040/ Original (Ukrainian): https://neurodrift.org/blog/robota-do-2040/ Tags: AI, jobs, economy, employment, scenarios Canaries were pulled from British coal mines back in 1986 — instruments had learned to detect the gas instead. In 2026, the canaries are back: employment for 22-to-25-year-olds in AI-exposed professions has dropped 16%, while US unemployment sits at a historic low. This is an autopsy of the central mistake in the whole AI-and-jobs debate: everyone's bracing for mass layoffs, while the real crisis arrives as a quiet non-invitation. Meet the infinite intern, find out why your salary is really rent on scarcity, learn who's already collecting the automation dividend (hint: the top 10% of Americans hold 87% of the stock market) — and walk through five scenarios for how this plays out by 2040. ----- British miners carried canaries underground for more than eighty years. A bird's heart beats too fast: where a person would still feel nothing, the canary has already dropped off its perch. They weren't officially retired from the mines until late 1986 — electronic sensors turned out cheaper, and Britain's two hundred state-owned canaries were pensioned off. No ceremony. The sensors were simply trusted more. In 2026, the canaries are back — this time in employment statistics. On paper, everything's fine. Three and a half years after ChatGPT launched, US unemployment sits at 4.2%, below the seventy-year average. The eurozone is near a historic low. Japan is at 2.6%, where employers are fighting over people. If you only look at these numbers, the panic around AI is just another cry of wolf. Now look where Stanford's researchers looked — into ADP's payroll records, the company that processes paychecks for millions of Americans. There you can see what the aggregate hides: since late 2022, employment for 22–25-year-olds in the most AI-exposed professions has dropped 16% relative to peers in protected professions. Among young software developers, it's down 20%. Meanwhile, people in their thirties in those same professions are up 6–12%. The researchers named their own paper for it: "Canaries in the Coal Mine." In Britain, graduate vacancies fell 45% in a single year. In the US, for the first time since measurement began, a basic axiom of the labor market broke: unemployment among fresh university graduates is now higher than for the workforce as a whole. The disruption didn't arrive as a layoff. A layoff is an event, a scandal, a headline. It arrived as a non-invitation — and a non-invitation never makes the news. The company doesn't fire anyone. It just quietly doesn't hire the person it would have hired in 2019. A canary at least got carried out of the mine into fresh air. The class of 2026 just doesn't get let into the mine at all — and they tell you the air's fine, because the sensors are silent. We won't lose our jobs. We'll lose the power that jobs gave us. The infinite intern Meet your new colleague. He doesn't sleep, doesn't get sick, doesn't ask for a raise, and doesn't fire off passive-aggressive messages in the team chat. He writes the first draft of the contract, the simple code, the boilerplate report — everything that, for forty years, made up a junior's first year on the job. He costs twenty dollars a month. And here's the key part: you don't hire him. You instantiate him. Ugly word, but accurate. The machine Luddites smashed in 1811 had to be cast, hauled in, and installed — one at a time. The ATM that was supposed to wipe out bank tellers went in one unit at a time. An industrial robot today still takes months to reach a factory floor. A trained agent gets copied a million times a minute. Marx called the unemployed a "reserve army of labor": their mere existence presses down on the wages of the employed. Now the army has gone synthetic — and infinite. Let's give the mechanism a name: the synthetic reserve army. A salary isn't payment for effort. It's rent on scarcity: you get paid exactly as much as it would cost to replace you. A lawyer bills four hundred dollars an hour not because the work is so grueling, but because there aren't many people with that particular head on their shoulders. The entire structure of the middle class — the mortgage, the kids' education, the pension — rests on the fact that expertise accumulates slowly and doesn't copy itself. AI is a technology for copying expertise. The pressure on your salary comes not from an actual replacement — but from its standing possibility. The employer doesn't need to actually buy a two-dollar-an-hour agent. It's enough for them to know one exists — for the moment you ask for a raise. The people building this army are the ones telling us what's coming next — and here it's worth pricing their words honestly. Dario Amodei, CEO of Anthropic: AI could wipe out half of entry-level white-collar jobs within one to five years, pushing unemployment in the affected group to 10–20%. Worth remembering: this is a warning from someone who sells exactly this technology — a forecast and a price list in the same breath. Sam Altman writes that society "will get so much wealthier so quickly" that it can seriously entertain ideas once considered unthinkable, while the future will look back on today's jobs as "very fake jobs." Ford's CEO Jim Farley — a man who keeps hundreds of thousands of workers on assembly lines — put it plainly about office workers: "AI's going to replace literally half of all white-collar workers in the US." Words are cheap. Expensive signals look different: Amazon warned in an internal memo that corporate headcount would shrink as agents roll out — then cut roughly thirty thousand office jobs within a few months. But the most honest signal came from Klarna: in 2024 it loudly replaced the work of seven hundred support agents with a bot; in 2025 it admitted quality had slipped — "we implemented AI too quickly to cut costs on the people side" — and brought humans back for the complicated conversations. Remember this one. It means two things at once: agents can already do a lot — and still can't do the thing that matters most. The employer doesn't need to actually buy a two-dollar-an-hour agent. It's enough for them to know one exists — for the moment you ask for a raise. The optics: 5%, 9%, 47%, or 80%? Arguments about AI and jobs are full of numbers that differ by a factor of sixteen — and every single one is "true." They're just answers to four different questions that keep getting mashed into one. 80% of American workers have at least one-tenth of their tasks touched by an LLM — OpenAI's count. That's about contact, not replacement. 47% of professions are "at high risk" — the famous 2013 Oxford estimate, which measured whole professions, as if a profession vanishes the moment its first task gets automated. 9% of jobs are realistically automatable — an OECD re-estimate that broke professions down task by task. And finally, 5% — that's how many tasks, according to Nobel laureate Daron Acemoglu, will be profitable to automate in the next decade, because there's a chasm between "technically possible" and "economically worthwhile once you account for errors and liability." Anyone juggling these numbers as if they're interchangeable either doesn't understand them or is selling you something. The same trick, inside a single bank. In 2023, Goldman Sachs put out the headline "300 million jobs at risk." That same Goldman, in 2025, measured not the "risk zone" but the actual displacement flow, and got: 15 million in the US over an entire decade. An order of magnitude went missing somewhere between the press release and the methodology. But there's one measure where the applause-line numbers don't lie — and it's scarier than any of them. A profession doesn't have to disappear for jobs to disappear. If a company used to employ a thousand office workers and the same volume of work now gets done by two hundred, the profession is technically still alive, and eight hundred chairs are empty. IBM automated 94% of its routine HR tasks — and grew its overall headcount: it redirected the freed-up money into engineers and salespeople. Meta reports a third more productivity per engineer — and holds a record 78,800 employees. Amazon, meanwhile, is cutting. The conclusion is uglier than either side's propaganda: layoffs are a management choice, not a technological inevitability. A growing firm converts productivity into a bigger product. A stalled firm converts it into a smaller headcount. Guess what the economy will do once growth runs out. The Engels' pause, take two The most honest historical parallel for the next fifteen years isn't "the Luddites were wrong, and so are you." It's an uglier one. England, 1780–1840. Output per worker rose 46%. Real wages rose 12%. For sixty years, machines enriched the economy while walking right past the people running them — economists call this the "Engels' pause," because this is the exact England he described in "The Condition of the Working Class." Then, from 1840 to 1900, wages caught up and overtook. Except "then" took two generations, and it didn't arrive on its own — it was delivered by unions, strikes, and an expanding franchise. In other words: political power, not the gratitude of factory owners. Now look at today's chart. Labor's share of US GDP has fallen for the forty-fifth year running: from 67–69% in the late seventies to 59–62% now. Corporate profits are at record highs. Union density is at a historic low. And when Chinese imports knocked out just over two million American jobs in the 2000s, the textbook version of "reallocation" never happened: research by Autor, Dorn, and Hanson found that the hardest-hit towns hadn't recovered even a decade later. The labor market isn't a spring. It's more like a bone: it heals, but crooked. Honesty demands a counterweight, and this one's serious. The most rigorous formal model of an economy under full automation — by Korinek and Trammell — doesn't predict a wage collapse as inevitable: the direction depends on returns to scale and on where technical progress gets pointed. ATMs actually increased the number of bank tellers at first: cheaper branches meant more branches. The radiologists Geoffrey Hinton advised people to stop training in 2016 are in short supply in 2026. By the strictest government measure, only 17–20% of American firms actually use AI in production; 95% of corporate pilots produced no measurable financial return; a Danish study found chatbot users saved a grand total of 2.8% of their working time. Two hundred years of automation panics have ended in new professions — 60% of today's jobs didn't exist in 1940. So the most honest framing is: slower than the visionaries are shouting. Faster than the economists are soothing. And nowhere near where the unemployment statistics are looking. The Engels' pause is exactly the "nothing bad happened" scenario: everyone's employed, the economy grows, and your share of it quietly dissolves, like sugar in someone else's tea. The Engels' pause is the "nothing bad" scenario: everyone's employed, the economy grows, and your share quietly dissolves, like sugar in someone else's tea. Who gets the dividend The defining question of this era is simple: who gets the money from AI-driven productivity — the worker, the consumer, the state, or the owner? Uncomfortable news: while we've been debating it, it's already been decided by default. The top 10% of American households own 87.2% of the stock market — up from 81% in 2013. The bottom half of the country owns 1.1%. Seven companies, which are essentially AI's infrastructure, account for roughly 30% of the entire S&P 500. Four hyperscalers have pushed planned capital spending to $725 billion for 2026 — the largest concentrated private investment program in human history. The automation dividend gets paid out every single day. Just not through a paycheck — through a brokerage account. And that account isn't yours. Now for the other side of the scale — the mechanisms that were supposed to redistribute that dividend. The European Parliament rejected a robot tax back in 2017. The "world's first Korean robot tax" is a media myth — that was actually a cut to a tax break. The windfall clause — the AI labs' lovely promise to share excess profits — has zero signatories. The scoreboard reads: zero enacted mechanisms anywhere on the planet for redistributing AI rent. History smirks a little here: the welfare states of the twentieth century weren't born from elite enlightenment — they were born from total war and full employment. The state needed millions of bodies on the front line and on the factory floor, and those bodies bargained themselves a share. The pension, incidentally, wasn't a gift either — it was won. It's the one form of unconditional income humanity has actually agreed to. You just have to live long enough to collect it. We won't lose our jobs. We'll lose the power that jobs gave us — because a strike only works while your hands are still needed. The demographic mask and China's laboratory The strongest argument against panic is, at the same time, the sneakiest trap. The developed world is aging fast: China will lose 239 million working-age people by 2050 — that's like erasing one and a half Americas from the labor market. Japan already lives in the future: third in the world for robot density and 2.6% unemployment, maximum automation alongside a chronic hunger for people. Korea has the planet's highest robot density and a birth rate of 0.75. Robots aren't coming to replace people. They're coming to replace people who were never going to exist in the first place. That sounds reassuring, right up until you notice the catch: demographics guarantee the unemployment rate will stay respectable under almost any scenario, because labor supply is contracting right along with labor demand. Politicians will look at their 4–5% and say there's no problem. The first thing artificial intelligence actually automated wasn't work — it was calm in the statistics. If you want a preview of the future, don't look at Silicon Valley — look at Beijing, the one place where both forces are colliding at full strength. On one side: an automation blitzkrieg — 54% of all industrial robot installations worldwide in a single year, over 97% of global humanoid robot shipments, a state fund behind it running around $139 billion. On the other side: youth unemployment that hit 21.3% in 2023, after which it wasn't brought down — it was redefined. The statistic got suspended for six months and relaunched under a new methodology, excluding students. An independent estimate from a Peking University economist, one that also counted people "lying flat," put the real number as high as 46.5%. The civil-service exam draws 98 applicants per opening. And here's the most interesting part: a journal published by the Communist Party's Central Party School is already calling, in plain language, to "scientifically regulate the pace of automation for the sake of employment stability," and a Beijing labor tribunal has already set a precedent that deploying AI is not lawful grounds for dismissal. The Communist Party became the first entity in the world to start manually throttling the speed of the future. Take a moment to admire it: the only institution that treated "a robot is cheaper than a person, but the unemployed are more dangerous than robots" as an engineering problem rather than a conference topic. Robots aren't coming to replace people. They're coming to replace people who were never going to exist in the first place — which is exactly why the statistics will stay quiet. Five scenarios for 2040 This isn't a menu the world picks one dish from. It's a palette — every region will mix its own combination. The probabilities are the author's own bet on each scenario becoming the dominant logic in at least one major region by 2040. | Scenario | Probability | What happens to employment | What happens to wages | Where it's most likely | |---|---|---|---|---| | 1. Absorption — the work keeps coming | ~25% | stable, composition shifts | rise, except at entry level | US, India | | 2. Time dividend — the 32-hour week | ~35% for the EU only | stable, fewer hours | flat | EU | | 3. Money dividend — UBI | 15% explicit / 80% creeping | decouples from income | transfers catch up | small rich countries | | 4. Rent — neo-feudal concentration | ~35–40% | officially fine | median flat, top rising | US | | 5. Human premium — care, education, live experience | ~70% as a component | flows into the "hug sector" | low without subsidy | everywhere | Absorption — the historical pattern holds for a fifth time: cheaper production creates new demand and new professions. That's what happened with farmers (41% of employment in 1900, 2% in 2000 — no collapse). Even in this gentlest scenario, the ladder doesn't get handed back to the young: the entry-level squeeze is already in the data. Time dividend — productivity gets converted not into money but into hours: the four-day week. Pilots have been brilliant (92% of companies in the British trial kept the new schedule), but history is against it: the US workweek hasn't gotten shorter since the 1950s, and when Belgium legislated the right to a compressed week, one percent of workers took it. People choose money for as long as money converts into status. Money dividend — an honest UBI at wage-replacement level costs around 21% of GDP, and no parliament on the planet has come anywhere close. The real form will be different — creeping: earlier pensions, wider benefits, longer "retraining stipends," until coverage becomes quasi-universal. UBI will arrive through the back door and go by a different name. Rent — the default scenario, because it isn't a prediction, it's an extrapolation: ownership concentration is already at record levels, redistribution mechanisms number zero, and labor's bargaining power keeps thinning out. Formally, nobody starves — there'll be enough cheap AI content and cheap calories for everyone. It's just that capital becomes economically sovereign, and labor becomes optional. Human premium — employment flows to where the human is the product: care, education, medicine, live experience. This is Baumol's law, and it works without fail — healthcare is already 17.8% of US GDP. The trap inside it: only the state can pay for the "hug sector" at scale, which means this scenario's fate depends on how workable scenario three turns out to be. And there's a shadow sixth scenario that never makes it into the forecasts, because it's already here. The theater of employment. Work performs four functions: income, status, a daily schedule, and social calm. Production is just one of them. India's MGNREGA program guarantees a hundred days of paid work a year to just over ninety million people — the largest employment factory in history. Japan spent the nineties flooding its crisis with "bridges to nowhere." China quotas graduate hiring into state companies. The West does the same thing, just shyly — compliance, licensing, approvals for approvals. In a world where the productive core can shrink to a minority of the employed, states will manufacture work, because the alternative is millions of young men with no schedule and no hierarchy. The question isn't whether this will happen. The question is whether we'll be honest enough to call it what it is. The bet for 2040: how this actually plays out After all the facts — here's a concrete bet, no hedging. There won't be mass unemployment: the aggregate rate holds in the 4–7% range, because demographics squeeze labor supply at roughly the same rate AI squeezes demand. Labor's share of GDP will fall faster than employment does — "a thousand becomes two hundred" stretches out into "a thousand becomes six hundred" over a decade, through attrition and non-hiring, without dramatic layoffs. The generation that's twenty right now will pay the most for the transition: a seniority crisis hits in 2033–2037, when it turns out nobody raised the juniors who were supposed to grow into the people managing the agents. The central political conflict won't be "robot versus human" — it'll be ownership of AI productivity. China will engineer a managed post-labor transition and call it stability. Europe will convert part of the dividend into time. America will convert it into concentration, with late transfers. And the physical world will outlast the office: a plumber will outlast a junior lawyer, because a plumber has something an agent doesn't — a body that can be sent out, and an insurance policy that can be sued. Scarcity won't disappear — it'll migrate. From intelligence into energy (the queue to connect to the US grid is already 2,600 gigawatts, with a wait measured in years), into frontier compute, and into three things that can't be copied: accountability, trust, and status. A human stays in the loop not because they're smarter than the model. But because they can be punished. A doctor, an auditor, a pilot — that's not just expertise; it's an address where accountability lives. Professions will hold up for exactly as long as their legal armor holds up. What's left is the hardest nut of all — status. In the largest unconditional-income experiment ever run (a thousand dollars a month, three years, three thousand participants), the effect on stress and mental health was significant in year one — and had completely vanished by year two, even though the money kept coming. Swedish lottery millionaires are more satisfied with life a decade later — but no happier day to day. And retirees, five to nine years after retiring, face a 30% higher suicide risk; after ten years, 47% higher. You can hand out money by decree. You can't hand out hierarchy that way: for someone to be above, someone has to be below. Work is a built-in clock, a social fabric, and a place in line for respect. The societies that come through this transition best won't be the ones that hand out the most money — they'll be the ones that learn to manufacture status outside of a paycheck: in caregiving, craft, sport, community. That sounds soft. It's the hardest engineering problem of the 2030s. Seven dashboard lights to track this by, for 2026–2030: whether the US graduate-unemployment inversion closes; whether labor's share of GDP breaks below 56%; whether real AI usage in production climbs past 30% of firms; whether even one G20 country passes a robot tax or a sovereign AI fund; whether the first legislated 32-hour week shows up in a major EU economy; whether humanoid robot shipments top a million a year; whether China formalizes "employment over automation" as official policy. Every light is a checkable claim. In five years, this piece can be judged against them. That's exactly how texts about the future should be judged. The canaries were retired in 1986 because instruments had learned to detect the gas. Our instruments — unemployment statistics — are calibrated for the last disaster. They're still glowing green while the air in the mine quietly changes composition: it isn't labor growing scarcer. It's the power labor used to carry. We won't lose our jobs. We'll lose the power that jobs gave us — unless we rewrite who owns the thing that's going to work in our place. The contract isn't signed yet. Not yet. Sources Brynjolfsson, Chandar, Chen — "Canaries in the Coal Mine?" (Stanford Digital Economy Lab, 2025): digitaleconomy.stanford.edu Acemoglu — "The Simple Macroeconomics of AI" (NBER WP 32487): nber.org/papers/w32487 Eloundou et al. — "GPTs are GPTs" (OpenAI, 2023): arxiv.org/abs/2303.10130 IMF — "Gen-AI: Artificial Intelligence and the Future of Work" (SDN/2024/001): imf.org Goldman Sachs Research — "The Potentially Large Effects of AI on Economic Growth" (2023): goldmansachs.com METR — "Time Horizon" (AI autonomous task horizon): metr.org/time-horizons Epoch AI — "LLM inference price trends": epoch.ai IFR — World Robotics 2025 (robot density and installations): ifr.org OpenResearch — Unconditional Income Study, employment effects (NBER w32719): nber.org/papers/w32719 Korinek, Trammell — "Economic Growth under Transformative AI" (NBER w31815): nber.org/papers/w31815 Brynjolfsson, Li, Raymond — "Generative AI at Work" (NBER w31161): nber.org/papers/w31161 ILO — Working Paper 140, GenAI and jobs (2025): ilo.org Rodrik — "Premature Deindustrialization" (NBER w20935): nber.org/w20935 Federal Reserve — Distributional Financial Accounts (distribution of stock ownership): federalreserve.gov IEA — "Energy and AI" (data centers and grids): iea.org Full research dossier (8 roles, ~90 sources, scenario matrix): in the author's archive; key primary sources are listed above.