TheJobsMarket
Automation and AI Exposure

Jobs Automation Has Reshaped Rather Than Removed

The teller is the case everyone cites and almost nobody finishes. There are still 347,400 of them.

Short answer

Automation more often changes what a job consists of than removes the job. Bank tellers are the standard example: cash machines took the cash handling and the occupation moved toward advice and account services, and there were still 347,400 tellers in 2024. The projections show a 12.9 per cent decline over the following decade — real, gradual, and nothing like disappearance.

The example everyone cites and nobody finishes

Cash machines were supposed to end bank tellers. There were still 347,400 of them in 2024, and the projections show a 12.9 per cent decline over the following decade — real, gradual, and nothing like disappearance.

What actually happened: the machine removed the single most repetitive task, branches got cheaper to run, banks opened more of them, and the role shifted toward selling and servicing products. Employment stayed high for decades before the slow decline began.

The prediction was not wrong about the technology. It was wrong about what a teller does.

Three things can follow from automating part of a job

The role narrows to what is left, and fewer people are needed. This is the outcome everybody assumes and it is not the most common one.

The role broadens, because the freed time gets spent on something more valuable. Headcount holds and the job description changes.

The service gets cheaper, demand rises, and more people are needed despite each doing less of the old work. This is what happened to tellers for thirty years.

Only the first reduces employment, and which one occurs depends on whether demand responds to price — which is a question about the market, not about the technology.

What changes even when headcount does not

The skill mix, and usually the wage distribution inside the occupation. When the routine part goes, what separates the top of an occupation from the bottom becomes the judgment part.

That tends to widen the gap between the tenth and ninetieth percentile within the same job title. The occupation survives, the median may hold, and the experience of being in it diverges sharply depending on which half you are in.

Why this is not the reassurance it sounds like

A job that survives in changed form still requires the person holding it to change. Somebody who was excellent at the automated part and indifferent at the rest is worse off, even though the occupation is stable and the statistics look fine.

Aggregate survival is not individual security. The aggregate is what gets quoted reassuringly, and it is the wrong unit for a decision about your own week.

The pattern across every case

The occupation keeps the part requiring judgment, accountability or a relationship, and sheds the part that was mechanical. Paralegals kept case management and lost document review. Radiologic technologists kept patient handling and clinical judgment. Travel agents kept the complex itinerary and lost the simple booking.

No exposure score can represent that, because a score attaches to a job title and the title is exactly what stays constant while everything underneath it moves.

The question to ask about your own role

If the most repetitive third of your week disappeared tomorrow, what would you spend it on, and would your employer consider that worth paying for?

A clear answer is a good sign and a plan. No answer is the actual risk, and it is worth finding one before the question is asked for you.

The version of this that is genuinely bad news

Reshaping is fine for an occupation and hard for a person, and the hardness is unevenly distributed. Somebody twenty-five years into a career built on the automated part has the least time to rebuild and the most to lose from doing so.

Employers rarely handle this well, because the reshaping happens gradually and nobody owns the transition. The person notices when the interesting work has quietly gone elsewhere, which is several years after the point at which acting would have been easy.

What an early move looks like

Not retraining into a different field. Moving within the same job toward the tasks that survive: the escalations, the non-standard cases, the client conversation, the decision somebody has to own.

Those are usually available to anybody who volunteers for them, cost nothing, and are far easier to take on while you still have the standing that comes from being good at the old work. That standing is the asset, and it depreciates.

Common questions

Did cash machines eliminate bank tellers?

No. There were still 347,400 in 2024, decades later, with a 12.9 per cent decline projected over the following ten years.

Why did teller employment hold up?

Branches became cheaper to run so banks opened more, and the role shifted toward advice and account services.

What are the possible outcomes of automating part of a job?

The role narrows, broadens into more valuable work, or expands because the cheaper service raises demand. Only the first cuts headcount.

Does a stable occupation mean I am safe?

No. Aggregate survival is not individual security — someone strong at the automated part and weak at the rest is worse off.

What question should I ask about my own role?

If the most repetitive third of your week vanished, what would you do with it, and would your employer pay for that?

Did cash machines eliminate bank tellers?

No. There were still 347,400 in 2024, with a 12.9 per cent decline projected over the following decade — decades after the technology arrived.

What are the possible outcomes of automating part of a job?

The role narrows, broadens into more valuable work, or expands because a cheaper service raises demand. Only the first cuts headcount.

Does a stable occupation mean I am safe?

No. Aggregate survival is not individual security — somebody strong at the automated part and weak at the rest is worse off while the statistics look fine.

CS

Charles Slocs

Data and research

Charles Slocs builds the data side of this site — pulling the federal wage and employment series, matching job titles to occupation codes, and working out what the numbers do and do not support. He writes the pages that are mostly a question about evidence: what a survey measured, how wide the spread really is, and which published figure is out of date.

All articles by Charles Slocs →