This series is adapted, with permission, from a Chinese-language series by our friend Marcus Ji, who looks at historical events through an economic lens. We have condensed and adapted the original for an international audience.
The oldest job in human history lasted roughly 1.79 million years. Yours might not last ten. And the reason will not be that AI does your job better than you – it will be the same reason that killed the oldest job ever recorded.
The oldest job: keeping the fire alive
Farming is about 10,000 years old. Scribes go back 5,000 years, blacksmiths 3,000. One job predates them all by an order of magnitude: fire-keeping – someone whose responsibility was to make sure the fire never went out.
The evidence is archaeological. Deep inside South Africa’s Wonderwerk Cave – 30 meters in, where no wildfire can physically reach – researchers found burnt animal bones in sediment layers dated between 1.07 and 1.79 million years old. The most plausible explanation: early Homo erectus captured fire from natural sources, carried it into the cave, and kept it burning. (Scholars debate when fire-keeping became a stable, transmitted role – some argue not until ~400,000 years ago. Even on the most conservative reading, it is vastly older than any other occupation.)
In the long ages before humans could make fire, an ember could only be taken from wildfire or lightning – and once it died, the next chance might be months away. Fire-keeping had every feature of a real job: a designated person, continuous duty, no interruptions tolerated, and catastrophic consequences for failure – the whole band could freeze or starve.
What actually killed it: the cost of rekindling went to zero
Rome kept a sacred flame, guarded by the Vestal Virgins, for roughly 1,100 years – and the story usually told is that a decree of Emperor Theodosius I extinguished it in 394 AD, one imperial order ending a job of a million years. But by then, ordinary Roman households owned fire-starting tools. The flame was a political symbol; the Vestals were guarding the fossil of a job, not a service anyone needed.
The job itself had been killed, in stages, by a technology chain: flint-and-pyrite fire-starting (~400,000 years ago) made rekindling possible but unreliable; portable fire kits – Ötzi the Iceman carried a complete one around 3300 BC – made it personal; and Iron Age fire steels (~1200 BC) made it trivial, about 30 seconds for a practiced hand. Once anyone could restart a fire at will, “watch the fire so it never dies” stopped being a need. The friction match of 1827 killed nothing – the job had been gone for three thousand years. (The reverse test holds: in the Andaman Islands, where fire-making techniques reportedly never arrived, communities were still maintaining permanent fires into the 20th century.)
Note what the Roman ending demonstrates: a job propped up by institutional power can be ended by a decree. A job that lives on real demand never dies by one.
This is the rule that governs everything below: jobs are not killed by better competitors. They are killed when technology removes the premise for their existence. The fire-keeper didn’t lose to a better fire-keeper.
The modern replays: nobody beat the middleman – the middleman got bypassed
The same mechanism, compressed from millennia into decades: US travel agents peaked at ~339,000 in 2000 and fell below 1% of that within about 20 years once airlines could sell directly to passengers. Typists peaked at ~900,000 in 1980 and were gone within 35 years of the IBM PC, once everyone typed for themselves. Telegraph operators lasted 68 years until the telephone made “encode the message and send it” an unnecessary step. None of them lost to stronger rivals; the need they sat in the middle of simply evaporated – exactly as it had for the spinners and weavers in the previous installment of this series.
The elevator operator: why “the technology isn’t good enough yet” is not a moat
One case proves the rule’s second half: technology removing the premise is necessary but not sufficient. There is one more variable – trust.
Automatic elevators existed by 1900. Yet in 1950, America still employed 114,473 elevator operators. The technology led to the job’s death by half a century, because passengers – even after Otis had solved precise floor-leveling in the 1940s – refused to step alone into an unmanned metal box.
The turn came on September 24, 1945, when some 15,000 New York elevator operators and building staff went on strike. About 1.5 million New Yorkers stayed home rather than climb stairs; federal tax collection reportedly lost $8 million a day. Building owners drew the obvious conclusion: this job was a systemic risk. Through the 1950s, automatic elevators were rolled out with a package of features engineered purely for passenger psychology — a big red emergency stop button (a sense of control), a voice announcement (“This is an automatic elevator…”), advertisements showing children and the elderly riding alone. By 1975 the occupation had effectively vanished. Harvard economist James Bessen traced all 270 occupations in the 1950 US census across the following 65 years and found exactly one that automation eliminated completely: elevator operator.
The lesson for anyone currently reassured that “AI still makes mistakes”: the coexistence period is not a safe zone – it is the window before trust finishes transferring. How long the window stays open depends on human psychology. That it will close depends on the technology, which only improves in one direction.
The order of extinction: white collar before cleaners
Line up the extinct occupations chronologically and an uncomfortable pattern appears: elevator operators went before cleaners, typists before waiters, telegraph operators before postmen. The more respectable, more trained job dies first.
The order is set by the return on replacement: labor cost saved, divided by technical difficulty. A clerk earns two or three times a cleaner’s wage, and clerical work – processing information by explicit rules – is precisely what machines copy most easily. A cleaner’s wage is low and the work environment is never the same twice, which machines find hard (roboticists call this Moravec’s paradox: what’s hard for humans is easy for machines, and vice versa). The highest-ROI target is always the job with the clearest rules, the most concentrated information processing, and the highest labor cost – which describes the white-collar middle, not the manual bottom. Scribes died of the printing press, telegraph operators of the telephone; junior document review, reconciliation work and template coding are the same line item in this century’s ledger.
What’s structurally different this time
Every generation of displaced workers believed in advance that they were safe, so “this time is different” deserves suspicion. But one difference is objective. Past machines replaced muscle, and the displaced moved into cognitive work — the skills lost and the skills needed next didn’t overlap. AI replaces cognition itself – reading, searching, summarising, drafting, analysing – so what’s being automated and what you’d retrain into overlap heavily. An analyst displaced by AI cannot escape by “learning analysis.”
And the burden falls unevenly. An NBER study tracked telephone operators displaced by automatic switching: ten years on, their employment rate was about 7 percentage points below comparable women their age. The hardest hit are never the young who haven’t entered the field – they are the mid-career, whose seniority devalues overnight.
The one-question self-test
All of the above compresses into a single testable line: could your job be written as an operations manual – given standard inputs, produce standard outputs – that someone else could follow? If yes, entirely: high exposure, regardless of how respectable the job looks. Mostly: medium. If the core value can’t be written down – judgment under incomplete information with your name on the outcome, hands-on work in unstructured physical and human situations, deciding what question to ask rather than answering it – you are safer, though not safe.
Don’t expect protection to change the outcome. The only “successful” resistance in history was the Ottoman Empire’s calligraphers, who – allied with religious authorities – suppressed Arabic and Turkish printing for roughly 250 years after Gutenberg. They saved two or three generations of livelihoods; the empire paid with two centuries of lag in knowledge transmission, missing the scientific revolution. Europe let its scribes die – and got publishers, editors, journalism and modern science in exchange.
The fire-keeper’s job survived from a South African cave to the Iron Age, and was ended not by an emperor’s decree but by a strip of steel that could strike a spark at will. The method for judging any job’s fate hasn’t changed in 1.79 million years: don’t ask how strong the competition is. Ask whether the premise for the job still exists.
Previously in this series: Why did the last mass job displacement take 200 years to complete?
Next in this series: How did a Guangzhou merchant become richer than the whole U.S. government?
