On a June day in 1967, a crowd gathered outside a Barclays branch in north London to watch a man feed a slip of paper into a beige box bolted to the wall. The box read it, thought for a moment, and pushed a crisp ten pound note into his hand. The bank's staff assured everyone that this was how people would get cash from now on.
The conclusion felt obvious to everyone watching: bank tellers were finished. Why would you need a human to hand you money through a window when a machine could do it around the clock?
Here is what actually happened. As ATMs spread, the number of bank tellers went up. It kept going up for decades, roughly doubling before peaking around 2007. The machine took over the cash-counting, and banks discovered that a branch staffed with humans who could sell mortgages, fix problems, and talk to customers was worth more than ever, and cheaper to open, so they opened more of them.
Every generation has this exact conversation. ATMs. Spreadsheets. Photography. The word processor. And now AI. So before you decide whether to panic, it is worth asking a more precise question.
What is a job, actually?
Strip away the titles and every job is the same shape: something goes in one side, something more valuable comes out the other. A nurse takes in a sick patient and outputs a healthier one. A writer takes in scattered information and outputs an article. A chef takes in raw ingredients and outputs dinner.
Between the input and the output sits a pile of tasks, and most of them are not the job. They are the grind the job currently requires. Junior consultants spend two or three years formatting slide decks through twenty rounds of revisions before anyone lets them near a client. Lawyers bill weeks to searching documents for the three pages that matter. Nobody grew up dreaming of doing either.
That distinction matters, because when people watch an AI do "their job," what they are almost always watching is an AI doing one of those tasks. The task is not the job. The job is getting the output the world wants from the input you were handed.
There is a classic survey question that makes the point brutally well: if you won 10 million dollars tomorrow, would you keep working? Across decades of asking, more than half of people say they would at least quit their current job. People do not love their tasks. They work to buy back their time. An assistant that eats the worst tasks first is not obviously your enemy.
A tale of two jobs
Two occupations tell you almost everything about how this plays out.
The elevator operator existed because early elevators were dangerous and manual. Someone had to work the lever, judge the speed, and stop level with the floor. Every elevator had a person in it for the better part of a century. Then automation arrived, and the job did not shrink. It vanished. In the US census's list of occupations, the elevator operator appears for the last time in 1960. It is the only job in the modern record that automation erased completely.
Now look at the airline pilot. Autopilot has existed since the 1920s. Fly-by-wire took over in the 1980s. Modern jets can land themselves in zero visibility. By the logic of the ATM crowd, pilots should be museum pieces. Instead, automation made every pilot more powerful, flying became cheap enough for hundreds of millions of ordinary people, and the number of pilots exploded. Because when a bird strikes the engine or a passenger collapses at 35,000 feet, the computer does not make the call. A trained human does, and every person on that plane wants them there.
The difference between the two is not how much of the work got automated. In both cases, nearly all of it did. The difference is whether there was anywhere for the human to grow. The elevator operator was sealed in a box with no new responsibilities to absorb. The pilot sat on top of an expanding industry with more planes, longer routes, and higher stakes. Automation shrank one job's box and blew the walls off the other's.
What 80 years of census data says
MIT economist David Autor and his team went through eight decades of US census records and traced every new job title as it appeared, from solar photovoltaic electrician to, yes, dog psychiatrist. The headline finding: the majority of jobs people work today did not exist a few generations ago. New work is invented constantly, and it appears fastest exactly where technology makes existing workers more valuable.
The pattern repeats everywhere you look. Spreadsheets did not eliminate accountants; they multiplied analysts. The nail gun did not end carpentry; houses got more elaborate. Word processors did not shrink newsrooms into nothing; there are vastly more working writers today than in the typewriter era. Digital editing did not kill film editing; it turned a physical splicing craft into an industry so productive that streaming became a global pastime.
And one more thing gets missed in every automation panic: most of what these tools create was never anyone's job to begin with. Economists call it competing against non-consumption. The custom part that was too expensive to manufacture, the analysis nobody had time to run, the design that would have taken a specialist a full day so it simply never got made. When the cost of making things collapses, the first thing that happens is not job theft. It is that thousands of things that used to be worth nothing to attempt suddenly get attempted.
Augmented or replaced: it is partly your call
So, will AI take your job? The honest answer from the history and the data: probably not. AI will take tasks, in some jobs almost all of the current tasks, and the job will reshape around what remains, which tends to be the judgment, the responsibility, and the human accountability that were the hard part all along.
But there is a sharper version of the answer, and it is the one worth acting on. AI probably will not take your job. Someone using AI will, if you do not.
Within a few years, refusing to work with AI will read the way refusing to use a computer reads today. And the gap compounds quietly: a person who squeezes even a 10 percent productivity gain out of these tools every week ends the year several times more productive than an identical colleague who never started. Not because the tools are magic. They are magic for about a week, and after that it is a learning curve like any other. The people who win are simply the ones who put in the reps.
That is the practical takeaway. Do not spend this decade guarding your tasks. Spend it becoming the pilot: the person automation makes more powerful, more trusted, and harder to replace.
If you want a structured way to get there, that is exactly what Mindwand is for: short, interactive lessons that make you genuinely good at using AI at work, one subscription for every course, with a prompt library to lean on. Start before your colleagues do.