The Map of Work After AI: Decomposing Professions into Tasks, the ATM Paradox, and the Failed Frey–Osborne Forecast Author: Дністер Published: 2026-09-16T03:01:41.000Z Language: en URL: https://neurodrift.org/en/blog/ai-ne-zabere-robotu-zabere-lyudyna-yaka-mysphenytt-zadachamy/ Original (Ukrainian): https://neurodrift.org/blog/ai-ne-zabere-robotu-zabere-lyudyna-yaka-mysphenytt-zadachamy/ Tags: ai, career, automation, skills, future-of-work, labor-market A longread about the labor market's most expensive illusion: that a «profession» is not your skills but the plaque on the door. AI doesn't kill professions head-on — it demobilizes them one task at a time, and you hand over the weapon yourself with the words «well, it's just routine». Through the story of Katherine Johnson, ATMs, John Henry and ten verified studies — a new career frame: verbs instead of nouns. ----- In February 1962 a man named John Glenn sat in a tin can the size of a phone booth, bolted to a rocket they had only just learned not to blow up on the launchpad. The capsule was called Friendship 7. The orbit it was meant to fly was computed by a room the size of a wardrobe — the IBM 7090, NASA's pride, a machine that cost as much as a suburban block. And just before launch Glenn gave the engineers one order that must have turned their stomachs cold: "Get the girl to check the numbers. If she says they're good, I'll go." The "girl" was the mathematician Katherine Johnson. She took a pencil and a mechanical calculator and spent a day and a half re-computing the machine's ENTIRE trajectory by hand, digit by digit. The numbers matched. Glenn flew. Sit with the chain of command in that moment: a million-dollar machine produced an answer — and that answer wasn't considered an answer until it had been signed off by a woman with a pencil, who at that same NASA wasn't allowed to pour coffee from the same pots as her white colleagues. Now look at the trap, because it is brilliant in its cruelty. The IBM 7090 didn't cancel Katherine Johnson. The machine took exactly ONE task from her — arithmetic — and by doing so promoted her to the task the hardware couldn't perform: being the person on whose word a frightened astronaut stakes his own life. Sixty years on, the lesson holds with a hard number in hand: a little over 60% of all the jobs Americans held as of 2018 were in occupations that appeared only after 1940 (Autor, Chin, Salomons and Seegmiller, QJE 2024 — four economists who took apart the US census by new occupations). Most of the work that sustains the country didn't even have a name in 1940. The computer didn't erase the human-computer. It erased the NOUN "computer" and rebuilt it into a VERB. That's what this whole text is about. And it isn't a consolation text. It's a text about who among us is Katherine Johnson, on whose word a life is staked — and who is John Henry, who right now is racing a steam drill with a hammer and still thinks that winning the race is survival. And so the question doesn't hang in a cosmic vacuum, let's lower it underground — to where it lands on the body. Picture a miner who spent thirty years cutting coal with a pick hammer and took pride in calluses that never healed. They bring in a shearer. The machine cuts in one shift what a crew used to take a week to do. And now two men with identical hands stand before the coalface. The first asks: "what else can these hands do — prop the roof, read the roof, hear the mountain grow heavy before a collapse?" — and climbs into the shearer's cab as an operator. The second asks: "who's better — me or it?", steps out with his hammer for a contest, and loses the war even if he wins the shift. This whole article is about how those two questions differ. Because the machine at the coalface is already humming. The lamp on your helmet still shines, for now. The only question is: are you reading the roof — or guarding the plaque "FACE No. 7" over a seam you mined out two quarters ago. The Antagonist Has a Face, and That Face Is Your Résumé Let's name the enemy, because a nameless enemy always wins. The enemy is called the Profession. It's a noun-fortress. It has a voice — confident, dignified, with a note of wounded pride — and that voice introduces itself by title: "I am a Copywriter with twelve years of experience." "I am an Accountant." "I am a Translator." The voice says it as if naming not a job but a soul. And like every soul, it sincerely believes it can't be fired — only betrayed. I know this voice from the inside, so I have the right to cut it. For half my life I introduced myself in nouns. Founder. Investor. At some point I caught myself defending not what I can DO, but the plaque on the door — and defending it from a machine that couldn't care less about plaques, because the machine can't read plaques. It can only read what a life is staked on, and what it isn't. The first, it asks of you. The second, it takes. And here's the antagonist's trick — it doesn't hit you in the face. It doesn't come and say "you're fired." It does something worse: the system rewards those who quietly disarm and call it maturity. It's the same internal PR department guarding not your skills but the signboard. It whispers: "Well, it's just routine. Let the machine do it. That's not what I trained for." And the specialist hands over the task with relief — first the boring one, then another, then another — until all that's left of the fortress is a signboard creaking in the wind over an empty building. And he stands beneath it and proudly repeats its name, like a guard watching over a safe from which everything was carried off in the night except the plaque "SAFE." This isn't a firing. It's a demobilization. A profession isn't killed by assault — it's disbanded task by task, and the market rewards those who sign their own demobilization order with the words "at least now I have more time for the strategic stuff." There really is more time. Of strategic stuff — zero, because the strategic stuff was the first thing handed over. The machine can't read the plaque on your door. It can only read what you actually do. And that is exactly why you're in danger to precisely the degree that your job is a noun with no verbs inside it. The State Already Disassembled You Into Parts. For Free Here someone will say: "Well, that's a pretty metaphor, task-verbs, but the profession is real." No. A profession is a marketing label stuck on top of a bundle of tasks. And this isn't my poetic license — it's the bookkeeping of the US Department of Labor. There's a database — ONET. Governmental. Boring. In it, 1,016 occupational titles are broken down into more than 19,000 statements of concrete tasks, which in turn are grouped into 2,000+ detailed work activities. Sit with it: the state has already done the most unpleasant work for you and for AI. It took every proud title, every "I am a specialist," laid it on the coroner's table and neatly separated it into organs — and then, unlike a real coroner, posted the autopsy report in open access so anyone curious could download exactly what you're made of. "SMM manager" turned out to be a little heap of a few dozen verbs: write, plan, calculate reach, reply, negotiate. And now the most important sentence in this text. AI doesn't shop on the floor of titles. It shops one floor down — on the floor of tasks. It doesn't buy "Translator." It buys the verb "translate." And while HR is still putting the noun-vacancy "Translator" in the window, AI on the neighboring market is already buying up the verb "translate" wholesale — and they're trading on two different exchanges, not even suspecting the other exists. While one exchange solemnly announces a vacancy, the other has already closed it in bulk and gone to lunch. This isn't a new panic of the ChatGPT era. This frame is twenty-three years old. Back in 2003 David Autor and colleagues laid out the cold mechanics: the computer REPLACES the human on ROUTINE tasks — where there's a clear rule — and COMPLEMENTS them on the non-routine: problem-solving, complex communication. Their model back then explained 60% of the demand shift toward educated labor over 1970–1998. And the unit of analysis was already the task. Always the task. Never the whole job. Twenty-three years they've drilled the same letter into us, and we still confidently write résumés in nouns. Why the Very ATM That Was Supposed to Kill the Teller Saved Him Instead If you still think automating a task = the death of a profession, here's the cleanest counterexample in history, and it's disgustingly simple. From the 1980s to 2010 the US installed about 400,000 ATMs. Any "common sense" said: tellers are finished, the machine counts money faster and doesn't steal. And now the number that makes common sense choke: the number of bank tellers over that same period did NOT fall, it ROSE — from roughly 500,000 to 600,000. The machine was installed literally to do the teller's job — and there were more tellers. If this were a detective story, the machine would have been released for lack of a corpse. How? The ATM made a branch cheaper to operate. A cheaper branch — banks opened MORE branches to fight for market share. And the teller's task moved up the stack: from "count the bills" to "sell a loan, retain the client, be the face." The machine ate the boring task — and the human migrated to the task where you have to look people in the eye. The teller didn't die. The teller moved up a floor. This is the pattern that repeats with the dumb regularity of the tide. The loom. The human-computer. The ATM. Now — the large language model. Each time the chorus wails "the profession is dead." Each time the one who moved up a floor survives. And each time the one who chained himself to a single task and called that chain his dignity perishes — and then engraves his job title on the chain too, just so they don't mix up the body. I saw that same ATM in its coal version long before the banks. In the Ruhr of the 1970s, when mechanized complexes appeared at the coalface, every collier faced exactly the same fork as the teller. Whoever asked "what else can these hands do" climbed into the shearer's cab — and his task moved up a floor: no longer hacking the seam but driving the machine, hearing the roof, catching the first crack of a collapse a second before it became an obituary. And whoever took pride with his whole gut in the fact that he "cuts coal by hand" kept his dignity — and lost his shift. Because dignity nailed to a single task descends in the cage along with the task. The machine didn't despise his craftsmanship. It simply carried out the one task around which he'd built his whole name — and the man rode down in the very same cage he once descended in with pride at the start of a shift. !A teller smiles and looks the client in the eye — the human task that survived; behind the glass the machines count bills, and an «employee» with a smooth faceless ovoid head endlessly feeds banknotes into the dispenser; a moth with a singed wing circles the lamp. Automate a task — and the human migrates up the stack. Chain the human to one task — and automation becomes his obituary. The difference between those two sentences is your entire career. The Expert Is Safe. The Beginner Is in the Crosshairs. And That's No Mistake Here most AI texts lie to you in a soothing direction: the machine will come for the "low-skilled," and the masters are safe. Reality is meaner and funnier — it came for your advantage first, precisely because you'd spent the longest fattening it. In the call center of one Fortune 500 company they deployed an AI assistant and honestly measured it across 5,179 agents (Brynjolfsson, Li, Raymond, NBER 2023). On AVERAGE the number of issues resolved per hour rose by 14%. But the average lies here, like the average salary in a country where one factory owner and a thousand of his workers are "on average" millionaires. Because for NOVICES and weak agents the gain was a full 34%, while for top performers the effect was minimal — effectively near zero (that's exactly how the authors describe it, qualitatively, not as a figure). Translated into human: AI didn't replace the expert. AI let the beginner PERFORM the expert's TASKS. Picture a junior who yesterday couldn't close a ticket without a prompt, and today, with ChatGPT under the desk, closes them at a senior's pace — and doesn't even blush, because he has no idea whose salary he just undercut. AI poured your twelve years of experience across the novices who showed up on Monday — and did it for free, while you paid for those twelve years with your own back. The moat around your "profession" doesn't leak from the top, where the masters sit — it leaks from the bottom, and it leaks fastest. The expertise you accumulated over years as a private fortress suddenly became a built-in feature, a free add-on to a twenty-dollar subscription. Your advantage didn't vanish — it became universally available, which for an advantage is the same as death. And here's the scale. OpenAI's own economists calculated: about 80% of American workers will see at least 10% of their tasks touched by AI, and about 19% will see more than half their tasks touched. But listen carefully to the figure everyone skips: only 1.8% of jobs have more than half their tasks under the hit of a "bare" language model. But add software on top — and that share jumps to 46%. That is, the threat lives at the task level and scales with the TOOL, not with the job title. Between "this barely touches me" and "half of me is automated" lies not a change of profession. It lies one plugin. Someone in the next room is installing it right now, and your safety has a shelf life that expires exactly when the install finishes. Counterargument: What if There's Simply No "Floor Above"? Let's stop here and do what most optimistic AI texts deliberately skip: honestly voice the strongest objection. The "teller-who-moved-up" pattern really does exist. But it assumes there's somewhere to move up to. That a floor above exists. That a person is capable of migrating cognitively — refocusing, retraining, reformulating their own tasks. Now picture a 55-year-old miner from the Dnipropetrovsk region or a warehouse worker in Bangladesh, whose "fortress" consists of physical tasks: lift, carry, stack, sort. The floor above, for him, isn't "move from routine to judgment." The floor above is a class of cognitive work, access to which requires education, language, a network, capital. He has none of those four in the necessary quantity. The ATM replaced the teller's task — and the teller moved to the task "look the client in the eye." But the Amazon warehouse robot displaced the packer — and the packer didn't "move to a higher-level task," because that task is in another country, in another language, and requires another diploma. The teller and the packer aren't one situation at different scales. They're different structural situations. Katherine Johnson is a brilliant metaphor. But she's also a woman with a university math education, in an institution where a floor above existed, even if she was barred from it by racism. Most people in the wave of automation aren't Katherine Johnson with blocked access. They're people without the math and without the floor an elevator could carry them up to. This objection has real force — and a text about thinking in tasks must look it in the eye, not walk past it. But here's what this objection doesn't explain: it concerns the frame "who survives the turbulence" — and the answer "not everyone" is true. However, the frame "noun vs. verb" is not advice about equal opportunity. It's a description of the market's mechanics: by what principle it displaces. To know the mechanics is at least to see where the wall actually stands, instead of running into it blind and being surprised. For some, this knowledge opens a door. For others, it merely explains why the door is shut and what needs to change at the level of the system, not the résumé. Who Pays the Most — and Whose Name Doesn't Appear in the WEF Report The figure "plus 78 million jobs by 2030" from the World Economic Forum report sounds soothing, until you ask: where exactly will those jobs appear — and where exactly will the 92 million that go minus disappear? Because those two flows don't intersect geographically, educationally, or demographically. The new places concentrate in the tech sectors of developed countries and among specialists with digital skills. The disappearing ones — in manufacturing, in low-skill services, in emerging markets. A textile factory in Vietnam where automation cancels 40,000 jobs — and a tech hub in Warsaw opening 4,000 jobs for data analysts. It's the same "shuffle of the deck" in the report and two different countries in reality. More concretely: the heaviest price is paid by manufacturing workers over 50 (and their chance of simply changing jobs is already minuscule — by OECD data, in any given year only about 6% of people aged 55–64 do so; not because they don't want to, but because the market looks at gray hair like a fracture in the roof), non-cognitive service roles (cleaners, cashiers, sorters — where the ATM pattern doesn't repeat, because the "floor above" either doesn't exist or demands a leap across an education chasm), and markets with young, poorly educated populations — Sub-Saharan Africa, South Asia — where automation destroys the very "path through the factory into the middle class" that Japan and Korea walked in the 20th century. The WEF's net balance of +78 million is the average temperature across the hospital, where in one ward the fever is 40 and in another there are cold sheets and a tag on the toe. The average doesn't heal — it merely strikes the balance between those who got better and those who've already been discharged downward. Why the Prophecy of 47% Dead Jobs Itself Died Let me show you what happens when you look at the world through noun-glasses instead of verb-glasses. In 2013 two Oxford researchers, Frey and Osborne, issued a prophecy that circled the planet: 47% of US jobs at high risk of computerization over roughly twenty years. Headlines screamed. Conferences shook. How did they count? They ranked WHOLE OCCUPATIONS by risk. Nouns. And the prophecy aged not merely badly — it aged exactly backwards, with a precision a broken clock would envy. What the noun-glasses predicted — and what life did (US, 2013→2021) Occupation Frey–Osborne forecast (2013) What actually happened (2013–2021) Insurance underwriters Among the HIGHEST automation risks Employment rose by 16.4% Recreational therapists Among the SAFEST Employment fell by 8.9% Read the table again and let it sink in. What was supposed to die first grew. What was supposed to outlive everyone shrank. The noun-glasses didn't just miss — they showed a compass needle pointing dead south when you needed north. And it's not because Frey and Osborne are fools — they're serious scholars. It's because the occupation level is precisely the level at which nothing can be predicted. Because a profession doesn't get automated. Its individual tasks get automated, and then what's left gets reshuffled into something new. To forecast the fate of a profession is to read tea leaves on the plaque on the door while behind it they're rearranging the room. The prophecy "47% of jobs will die" died precisely because it looked at nouns. The underwriter was supposed to die — and grew. The therapist was supposed to survive — and shrank. A compass pointing south isn't broken. It's just measuring the wrong thing. The Market Doesn't Die. It Shuffles the Deck — and You're Still Clutching One Card Let's bring the big numbers together, because they paint a single picture, and that picture isn't a funeral but a delivery room with elements of a slaughterhouse. The World Economic Forum, having surveyed over a thousand employers and drawing on data about more than 14 million workers: by 2030, 170 million new jobs will be created, 92 million will disappear, net gain — plus 78 million. Total "churn" — 22% of all jobs. Goldman Sachs: about 300 million full-time equivalents globally under some automation, two-thirds of jobs in the US and Europe partially exposed — but precisely PARTIALLY, by a chunk of tasks, not wholesale. And a fresh McKinsey from November 2025 throws the crudest figure on the table: it's already technically possible today to automate 57% of work HOURS in the US. 57%. More than half the hours. And here's where McKinsey does what the noun-glasses didn't — it warns outright: this is "technical potential in TASKS, not an inevitable loss of jobs." Over 70% of the skills employers seek today are used in both automated and non-automated roles simultaneously. The machine eats hours. The machine eats tasks. And the job survives by reshuffling the remaining tasks into a new configuration. The deck isn't burned. The deck is dealt anew — and to rise from the table with the words "I always played this card" is to lose before the deal is even done. The labor market doesn't shrink — it brutally reshuffles the deck of tasks, and the loser isn't the one with a «dangerous profession», but the one who refuses to continuously re-trade their own set of tasks against the machine. A person with a profession-frame experiences this shuffle as an existential death: "my profession is gone, therefore I'm gone." A person with a task-frame experiences the very same thing as a tool upgrade: "oh, the machine does this task now, my hands are freed up — where do I redirect them?" The same event. Two opposite emotions. The difference is only in what you glued yourself to: the noun or the verb. The first writes an obituary. The second — a shopping list. While You Defend the Noun, the Company Has Already Switched to Verbs So far we've stood over a single person with a sheet of paper. But there's a floor above your anxiety — and on it the same reconfiguration proceeds coldly, with no existential drama, like the planned closure of a longwall. While you defend yourself with "I'm a Copywriter," your employer has already quietly stopped thinking in your job title. He thinks in your tasks — and did so long before he uttered the abbreviation AI. It's called by the boring term skills-based organization, and behind it stands simple surgery: the company breaks the job down into the very bundle of verbs ONET uses, then puts them back together differently — handing part to the machine, part to a freelance agent on a platform, part stitched into a new role that doesn't yet have a name on the labor market. That same McKinsey that gave us 57% of hours puts it plainly: over 70% of the skills employers seek today migrate between automated and live roles simultaneously. Translation: HR has long been buying not you but your stockpile of verbs — it just kept writing your title on the price tag, because it's more convenient to pay one salary than to count each task separately. AI took that convenience away. Now every verb is visible piece by piece — and is traded piece by piece too. And here's where this cold machinery becomes personal. The manager doesn't look at your résumé or weigh your dignity — he does exactly what the mine director did with mechanization: he overlays the shearer onto the layout of the seam and counts how many colliers to leave in the cabs and how many — up top, in the HR department, where the severance pay already waits. The company doesn't fire a "profession." It rewrites the lines in the task matrix one by one — and your job disappears not with a gunshot but with a verb being quietly pulled out from under it, verb by verb, until the plaque hangs over an empty desk. The only difference is that you can draw up this matrix on yourself before HR does. In fact, the next section is about how to snatch the pencil from them. The Tool: An Audit of Your Own Task Stack, Glenn-in-Reverse Enough diagnoses. Here's a scalpel for your hand. This is a checklist that kills the antagonist with a single action — it asks John Glenn's question in reverse. Glenn asked "who is the person I'll trust my life to?" You ask about yourself: "which exact tasks can I right now — the only one in the room — close in a way that a life would be staked on them?" Take a sheet. Break your job into 7–10 concrete VERBS — what you physically do over a week (not "I'm a marketer" but "I write copy," "I analyze numbers," "I negotiate with a contractor," "I invent hypotheses"). Then for each one honestly mark: The task-stack matrix: verb versus machine Task (verb) Does the machine already do this? Is it routine or judgment/trust? Would I stake a life on it? Action E.g.: draft the text Yes, well Routine No Hand it to the machine CONSCIOUSLY, take the freed-up time E.g.: decide what here is true No / unreliable Judgment Yes Deepen it — this is your floor above E.g.: negotiate under pressure No Trust Yes This is your arithmetic-in-reverse: what was left to Katherine The rule for reading the matrix is cruel and simple. Every row with "would I stake a life on it? — NO" and "routine" is not your dignity. It's waste rock in the cart you're still hauling up because your title is stamped on it. Hand it to the machine CONSCIOUSLY — that's different from having it taken. Whoever empties the cart himself keeps the freed-up hands for himself. Whoever waits for it to be emptied keeps for himself only a panic attack and a sick note. The same rock. The difference is in who holds the shovel — you or the HR department. And the rows with "stake a life? — YES" are your real front. This is what Glenn called the girl for. If you have zero such rows — you have no profession. You have a signboard. And that's the most important — and most unpleasant — thing you'll learn about yourself this week: maybe for the first time you'll look at your own job and find under the plaque not a single number you'd stake a life on. John Henry Won the Fight and Lost the War I'll end not with consolation. I'll end with a grave. Here we have to honestly separate the document from the legend, because mixing them is the very sin of the noun. The document is this: the historian Scott Reynolds Nelson found the real John Henry in Virginia's prison archives — a 19-year-old Black prisoner, convicted in 1866 and thrown onto building the Chesapeake & Ohio (C&O) railroad among other inmates, alongside steam drills that worked day and night. That's a fact from the papers. And from there begins the ballad sung by generations, whose numbers you should trust as a song, not as a record: that the steel-driver John Henry went out to race a machine, drilled more than it with a bare hammer — and fell dead with the hammer in his hand, having "won" the race. The document gives us a body. The ballad gives us a moral. And the moral here is scarier than any record. Because John Henry defined himself through one single task: hammering steel. And when the machine that did the same was brought in, he didn't ask "what else can I do?" He took the fight on the same floor — and it was precisely the victory that killed him. Here's the antipode of Katherine Johnson in two sentences. Johnson handed the machine the arithmetic and rose to the task "be the one they trust with their lives" — and lived to 101 as a legend whom decades later they'd play in a film. Henry clung to a single task, called it honor, won the round — and it killed him, and the reward was a song about how beautifully he lost. Both met the machine. One asked "what else can I do?" The other — "who's better, me or it?" The first question is about verbs. The second — about the noun. And it's the second question that's on Henry's headstone. !An exhausted John Henry in a work shirt stands at attention with his hammer beside an empty desk, like a guard over a safe emptied of everything; in the doorway behind him — a bare hall where under a spotlight a single rubber stamp remains on a pedestal, and a transparent janitor-function drags a mop that touches nothing; on the desk lies a dead moth with burned wings. He stands guard over a job whose tasks were carried off years ago, and he mistakes standing guard for victory. AI won't take your job. That's true, though not the truth you were hoping for. Your job will be taken by the person who stopped being a noun and learned to think in verbs — while you heroically swing the hammer, proving to the machine that you're still a Copywriter. The machine won't come for the plaque on your door. For your plaque comes a colleague who took his own off first — the same Ruhr collier who didn't compete with the shearer but sat in its cab. While you shook the machine's hand in a sign of fair single combat, he'd already grabbed its levers — and rode up a floor in the very cage you're still waiting for down below. Glenn's question still hangs in the air, only now it's asked not of an astronaut but of you. Not "what do you do for work." But: which exact numbers can you right now — the only one in the room — re-compute in a way that a life would be staked on them? If there's an answer — rise from the table and go defend it before it gets digitized. If there's no answer — you're not in a crisis. You have a day and a half and a pencil. Count. Sources Autor, Chin, Salomons, Seegmiller, «New Frontiers: The Origins and Content of New Work, 1940–2018», QJE 139(3), August 2024 (a little over 60% of 2018 employment in occupations introduced after 1940; phrasing «majority… slightly above 60%») — https://academic.oup.com/qje/article-abstract/139/3/1399/7630187 Autor, Levy, Murnane, «The Skill Content of Recent Technological Change», QJE 118(4), November 2003 (the model explains 60% of the demand shift toward educated labor, 1970–1998) — https://economics.mit.edu/sites/default/files/publications/the%20skill%20content%202003.pdf Eloundou, Manning, Mishkin, Rock, «GPTs are GPTs», arXiv 2303.10130 / Science 2024 (80% / 10% of tasks; 19% / more than half; 1.8% for a «bare» LLM versus ~46% with software) — https://arxiv.org/abs/2303.10130 McKinsey Global Institute, «Agents, Robots, and Us: Skill Partnerships in the Age of AI», 25 November 2025 (57% of work hours — technical automation potential: AI agents 44%, robots 13%; explicitly as task-level potential, not a forecast of job loss; over 70% of skills shared) — https://www.mckinsey.com/mgi/our-research/agents-robots-and-us-skill-partnerships-in-the-age-of-ai World Economic Forum, Future of Jobs Report 2025, January 2025 (+170M / −92M / +78M net; 22% churn; 1000+ employers surveyed, data on 14M+ workers) — https://www.weforum.org/publications/the-future-of-jobs-report-2025/ James Bessen, «Toil and Technology», IMF Finance & Development, March 2015 (≈400,000 ATMs 1980s–2010; tellers ≈500,000 → ≈600,000) — https://www.imf.org/external/pubs/ft/fandd/2015/03/bessen.htm Brynjolfsson, Li, Raymond, «Generative AI at Work», NBER WP 31161, April 2023 (5,179 agents; +14% on average, +34% for novices; for top performers the effect is described qualitatively as minimal / near zero, with no separate figure) — https://www.nber.org/papers/w31161 Frey & Osborne, «The Future of Employment» (2013); re-evaluation via ITIF, September 2022 (underwriters +16.4%, recreational therapists −8.9%, 2013–2021) — https://itif.org/publications/2022/09/30/oops-the-predicted-47-percent-of-job-loss-from-ai-didnt-happen/ Goldman Sachs Research, «Generative AI could raise global GDP by 7%», March 2023 (≈300M full-time equivalents affected; two-thirds of jobs in the US/Europe partially exposed; via CNBC) — https://www.cnbc.com/2023/03/28/ai-automation-could-impact-300-million-jobs-heres-which-ones.html O*NET Resource Center (US Department of Labor), database documentation (1,016 occupational titles; 19,000+ task statements; 2,000+ detailed work activities) — https://www.onetcenter.org/database.html John Glenn and Katherine Johnson, Friendship 7 (1962): «get the girl», manual trajectory verification over ~1.5–3 days; Johnson died 24 February 2020 at age 101 — Space.com / NASA — https://www.space.com/35218-hidden-figures-when-did-john-glenn-ask-for-the-girl-to-check-the-numbers.html The legend of John Henry: a documented person (19-year-old prisoner, 1866, C&O railroad) per Scott Reynolds Nelson, «Steel Drivin' Man» (2006); the details of the contest with the steam drill and the «victory at the cost of his life» belong to the ballad, not the archive — https://en.wikipedia.org/wiki/JohnHenry(folklore) OECD, «Promoting Better Career Choices for Longer Working Lives», March 2024 (only ≈6% of workers aged 55–64 change jobs in any given year) — https://www.oecd.org/en/publications/promoting-better-career-choices-for-longer-working-lives_1ef9a0d0-en.html