▲ 530 points
back
583 comments
One key line about ATMs is buried deep in the article:
> the number of tellers per branch fell by more than a third between 1988 and 2004, but the number of urban bank branches (also encouraged by a wave of bank deregulation allowing more branches) rose by more than 40 percent
So, ATMs did impact bank teller jobs by a significant amount. A third of them were made redundant. It's just that the decrease at individual bank branches was offset by the increase in the total number of branches, because of deregulation and a booming economy and whatever else.
A lot of AI predictions are based on the same premise. That AI will impact the economy in certain sectors, but the productivity gains will create new jobs and grow the size of the pie and we will all benefit.
But will it?
> But will it?
My prediction is no, because productivity gains must benefit the lower classes to see a multiplier in the economy.
For example, ATMs being automated did cause a negative drop in teller jobs, but fast money any time does increase the velocity of money in the economy. It decreases savings rate and encourages spending among the class of people whose money imparts the highest multiplier.
AI does not. All the spending on AI goes to a very small minority, who have a high savings rate. Junior employees that would have productively joined the labor force at good wages, must now compete to join the labor force at lower wages, depressing their purchasing power and reducing the flow of money.
Look at all the most used things for AI: cutting out menial decisions such as customer service. There are no "productivity" gains for the economy here. Each person in the US hired to do that job would spend their entire paycheck. Now instead, that money goes to a mega-corp and the savings is passed on to execs. The price of the service provided is not dropping (yet). Thus, no technology savings is occurring, either.
In my mind, the outcomes are:
* Lower quality services
* Higher savings rate
* K-shaped economy catering to the high earners
* Sticky prices
* Concentration of compute in AI companies
* Increased price of compute prevents new entrants from utilizing AI without paying rent-seekers, the AI companies
* Cycle continues all previous steps
We may reach a point where the only ones able to afford compute are AI companies and those that can pay AI companies. Where is the innovation then? It is a unique failure outcome I have yet to see anyone talk about, even though the supply and demand issues are present right now.
IIRC, the way this worked was that by decreasing tellers required per branch, it made a lot more marginal locations pencil out for branches, at a time when the banking industry was expansionary.
This is not so helpful if AI is boosting productivity while a sector is slowing down, because companies will cut in an overabundant market where deflationary pressure exists.
We're already seeing large software companies figure out that they don't need 5,000 developers. They probably only need 1,000 or maybe even fewer.
However, the number of software companies being started is booming which should result in net neutral or net positive in software developer employment.
Today: 100 software companies employ 1,000 developers each[0]
Tomorrow: 10,000 software companies employ 10 developers each[1]
The net is the same.
[0]https://x.com/jack/status/2027129697092731343
[1]https://www.linkedin.com/news/story/entrepreneurial-spirit-s...
> A third of them were made redundant
If I'm reading this correctly, the interpretation should be that a third of them were transferred to new branches.
0.66 (two thirds retention) * 1.4 (40% more branches) = 0.84, so we only expect ~16% were made redundant.
Correct. The story isn’t correct even in the original formulation. US population increased by 50% from 1980 to 2010, and the economy became far more financialized. But the number of bank teller jobs barely grew during that period, even before the iPhone.
I go back and forth on this. I relate it to software. I don't think AI can meaningfully write software autonomously. There are people who oversee it and prompt it and even then it might write things badly. So there needs to be a person in the loop. But that person should probably have very deep knowledge of the software especially for say low level coding. But then that person probably developed the knowledge by coding things by hand for a long time. Coding things by hand is part of getting the knowledge. But people especially students rely heavily on AI to write code so I assume their knowledge growth is stunted. I don't know mathematical proofs will help here. The specs have to come from somewhere.
I can see AI making things more productive but it requires humans to be very expert and do more work. That might mean fewer developers but they are all more skilled. It will take a while for people to level up so to speak. It's hard to predict but I think there could be a rough transition period because people haven't caught on that they can't rely on AI so either they will have to get a new career or ironically study harder.
> So, ATMs did impact bank teller jobs by a significant amount.
Did it? This sounds like describing a company opening a new campus as laying off a third of their employees, partly offset by most of them still having the same job in the same company but at a new desk.
> A third of them were made redundant.
More like something closer to 100%. The ATM was notable for enabling a complete change in mission. The historical job of teller largely disappeared, but a brand new job never done before was created in its wake. That is why there was little change in the number of people employed.
> because of deregulation and a booming economy and whatever else.
The deregulation largely happened in the 1970s, while you're talking about 1988 onward. The reality is that ATM actually was the primary catalyst for the specific branch expansion you are talking about. Like above, the ATM made the job of teller redundant, but it introduced a brand new job. A job that was most effective when the workers were closer to the customer, hence why workers were relocated.
It costs a lot of money to train one person to learn stuff.
We are already now in the time were training one LLM seems to be more cost effective to train for everything than training a million people the same thing over and over (after all, people loose knowledge when they get replaced).
LLM don't even need to become AGI to continue this trend. They just need to be good enough 'executors' of these tasks we expected people to do.
Which also means that every new job, which needs any form of training, will not be created because we will train ONE llm (or three, doesn't matter) to do it right and again you optimized the new people away.
Two anecdotes I'll share:
First: Most people believe it was Netflix that killed Blockbuster, but that's not strictly correct. It was the combination of Netflix and Redbox that really sealed the deal for Blockbuster (and video rental generally). It normally takes not one, but at least two things to really fill the full functionality of a old paradigm. Also it's human nature to focus heavily on one thing (Blockbuster was aware of Netflix) but lose sight of getting flanked by something else.
Second: Not listed here is how banks themselves have changed to be almost entirely online, which in many cases is more of a outsourcing play than a labor destruction play. My favorite example of this is Capital One, where the vast majority of their credit card operations literally cannot be solved in a branch. You must call them to say, resolve a fraud dispute. Note that this still requires staffing and is (not yet) fully automated, just not branch staffing. It doesn't make sense to staff branches to do that.
I do not get what's special about banking apps as opposed to online banking. I've been doing online banking in the browser on a PC since before apps and I'm still doing it because dealing with data on a phone is painful compared to a PC.
Is an app really that much easier to use?
Nice try, but no. When I was working for a US bank in the 80's, well before smartphone and even well before mobile phones the plan was hatched to reduce the number of offices because those offices were horrendously expensive. The big cost was the tellers and the handling of cash. For mortgages and other big ticket items there was a profit, but everything involved in the handling of money was a really large cost.
So they decided to reduce the number of offices. The ATMs were very specifically placed in the same location where the closed offices were, often renting just a fraction of the former space (usually a small cubbyhole attached to an outer wall). From 140 branches over a really small area they went to a small fraction of that, and ATMs took up the slack. Many people even preferred dealing with the ATMs rather than with the tellers because the ATMs were (at least initially) open 24x7.
Bank offices have all but disappeared. I think there are still two regional centers here and that's it. All deposits and all withdrawals of cash - as long as we still have cash - is handled by the ATMs. The iPhone came decades later.
TFA reasonably reduces to:
First, ATMs increased the demand for bank branches, which more than made up for the decrease in tellers per branch.
Second, mobile banking decreased the demand for physical branches.
I didn't notice any link with the iPhone, except maybe a vague coincidence in timing. Online banking existed before the iPhone, it worked using websites, on personal computers. And it took some time before smartphones were taken seriously by banks.
What I noticed however is a noticeable decrease in service quality in bank branches while online (desktop browser) options became better. Banks pushed customers out of their branches progressively. In the early 2010s tellers couldn't do anything you couldn't do online by yourself. For services like dealing with large quantities of cash, or coins, they made it so that you couldn't do more than what the ATMs allowed you to do, limiting the amount of cash the branch had access to and increasing how much you could withdrew from ATMs.
They didn't get the idea to fire all their tellers when Steve Jobs announced the iPhone. It was a decision at least a decade in the making. It is just that people tend to resist change so it happens slowly, especially for big, serious business like banking. And I don't think it is a bad thing.
I'm based in the rich Western world. Whenever I travel elsewhere, I'm amazed by the cheapness of labor.
Humans would attend a gas station or fetch items in a store. Why? They're completely unneeded, I can do (and WANT to do) that myself.
I always feel sad about these people, trapped in an economic system that forces them into useless labour when they could spend their time learning actually useful skills.
I remember a few years back when I went into my banking branch to deposit a check my insurance company had sent me. When it was my turn at the teller, and I presented my check to her for deposit, and she said, rather rudely: "You can do this with your phone, you know?" On my way out of the bank, I remember thinking to myself that she was, essentially, putting herself out of a job by encouraging people to use their phones and not her. Turns out I was correct.
That paired with an increasingly cashless society. (Which is also in large part to smart phones) Otherwise you'd still need more tellers to conduct transactions that exceed ATM limits.
Most responses here are reacting to the specifics of ATMs and bank tellers, but I think the more interesting point, which seems to be the point of the story, is that paradigm shifts (e.g at-home vs at-the-bank banking) can be more disruptive than automation.
The interesting question of course is what paradigm shifts may be enabled by AI? Certainly all the use case emphasis so far has been on automation, whether that's businesses using agentic workflows to replace manual ones, or agentic coding tools to automate the coding (and to much less degree software engineering) process. So far it's all mechanical horses.
For example, maybe (I don't see it, but maybe) the need for software goes away entirely since it's just an intermediary to getting something done. What if the AI can just do things for you directly, given specific instructions? Rather than giving detailed instructions to an AI to help you code some software, you (or someone/something) instead just bypass that step and give it detailed instructions to do whatever the software would have been used to accomplish.
As another off the top of my head example, what about healthcare? Are doctors and doctors offices the tellers and banks? We need to advance from brittle LLMs to robust AGI first, but at-home diagnosis and prescription could certainly replace many routine doctors office visits.
In recent years I have been going less and less to banks. 20 years ago I would go monthly to pay some bills.
Nowadays, I must visit a bank once or twice a year tops. My manager frequently sends me messages, but invariably he is trying to sell me something.
I've noticed that branches have really cut down on tellers and in my latest visit the branch didn't even have a teller, just someone helping people use the ATM and lots of desks (most were empty) for you to handle more complicated business with your account manager.
I've been thinking hard about this paradigm shift while thinking about ideas for things to "vibecode"
I started by trying to think about ways of running a vending machine company autonomously using a finite state machine + agents. It turns out most of "automating" a vending machine company doesn't need LLM agents at all, and simply buying machines with reliable telemetry + a database + automated inventory could get you much further than replacing every or even some components with an LLM. The LLM could replace the person on the phone texting the laborers who refill and service the machines, perhaps autonomously order refills (but hey so can a cronjob).
The troubling thought I had is that AI does not displace the technicians, or the vending machines. It replaces the manager. The human manager is the component that is unnecessary. The entire global economy can eventually reflect this reality where most of the wealth is technically owned by humans but where the majority of financial transactions and decision making will be done by machines (at a level not yet seen)
Macroeconomic metrics will go up along with wealth and standard of living, but for actual flesh and blood humans, much of this will be irrelevant.
This is a whole wall of post hoc ergo propter hoc.
You can’t state with any certainty that the ATM’s increased efficiency had anything to do with the expansion of bank branches. That could have simply been due to the strong population and economic growth. It’s quite possible (and I’d assume it to be true) that if the ATM had never been invented, there would have been far more bank tellers in 2005 than there were.
You also can’t assume the iPhone had that much to do with it. With the exception of depositing checks, there was nothing I couldn’t do on my computer in 2005 that I could on my phone in 2025. And you could always deposit a check at an ATM. It wasn’t like in 2006 we were all like “well I can only check my bank balance on my laptop so I’m going to drive there instead.”
It seems quite likely that other trends caused all of this.
I feel like even the phrasing of the original assumption that "we have more bank tellers now that we had before", which seems to imply that ATMs didn't affect or even boost the number of bank tellers is flawed.
If you look at the graph, the number of bank tellers from 1980 to 2010 went from roughly 500k to 550k (a 10% increase). However, the U.S. population grew from 220M to 305M in the same period (a 40% increase). To me, that seems to indicate that less and less people were becoming bank tellers after the invention of the ATM. Although from the graph again, you can see that the correlation is quite poor anyway.
Fun story. There are still bank tellers in the Falkland Islands because there is no e-banking. Transfers are literally made by filling in a piece of paper and taking it to the bank.
When ATMs first came out, they were mostly still only at the branch because they were big machines. I remember in the late 70s/early 80s, if you got a steady check (like social security or a paycheck from a steady job) you could cash them at the liquor store. The liquor store would even run my Dad a tab, and he would pay it off when he cashed the check. On paydays he would not be the only one doing that, they must have had to get a lot of cash on hand.
Starting with quotes with JD Vance and talking about listening to him on Joe Rogen is... a choice. Also I fail to see how the iPhone did anything or is relevant at all. Banking apps were made by third parties years after the iPhone came out and everybody had dozens of smart phones to choose from. The reason why they mentioned the iPhone specifically, touch screen and app store, already existed in the form of PDAs long before the iPhone came out.
ATMs came in the 90s, online banking in the 2000s, banks closed most of their branch offices in the 2010s i think. Gradually cash disappeared, so now you don't have ATMs either anymore. Than after covid they discovered that even the final bit of financial consultancy could be done via zoom, online.
Banking apps came later, long after banks had moved most interaction online.
> an AI system is literally a machine that can think and do things itself
why do so many writers claim this as a matter of fact? are we losing (or did we never have) a shared definition of the word "think"? can an LLM, at this time, function with zero human input whatsoever?
edit to add: these are genuine questions, not meant to be rhetorical :)
it's hard for me to gauge a broader understanding of AI/LLMs since most of the conversations i experience around them are here, or in negative contexts with people i know. and i'll admit i'm one of those negative people, but my general aversion to AI mostly has to do with my own anxiety around my mental health and cognitive ability in a use-it-or-lose-it sense, along with a disdain for its use in traditionally-creative fields.
From my experience, the banks did kill the teller jobs to save few pennies on the dollar. The result, here, is a very poor service compared to what we had in the past. I have witnessed very sad, inhumane and awkward situations in Greek banks.
Domestic appliances killed domestic service jobs.
Telephones killed messenger-boy jobs.
The automatic telephone exchange killed telephone operator jobs.
Movable-type presses killed the job of scribes despite the huge expansion in book production.
Various farm machines together killed arable farm labour.
The Laserjet and Wang word processor killed typist jobs.
Mainframe computers killed invoicing clerk, general accounting clerk, and inventory control clerk jobs.
We could go on.
In each case, the minor tasks in each job that were not automated were just folded into other jobs.
Focusing on ATMs and claiming no impact is egregious, tendentious cherry-picking. Machines almost always eliminate occupations.
If anyone who likes to geek out on old hardware, the image of the teller shows hardware that is part of the IBM 4700 Finance Communication System released in 1982, specifically the IBM 4704 Teller Terminal:
Archived docs:
https://bitsavers.org/pdf/ibm/4700/
Article on the history of ATMs:
https://computer.rip/2026-02-27-ibm-atm.html
(ChatGPT was of no use figuring this out)
I think people sometimes forget how backwards the US is, when lived in SF 7 years ago, you couldn't do wire transfers online. Maybe some banks, maybe some people.
But I constantly had issues with debit cards being rejected, wire transfers having to be done on a branch, etc. I doubt there is a modern bill payment system yet.
Where as in Denmark, I've bought house, mortgage, wired >100k, bought stonks, none of it required me going to a branch.
I pay a manual bill maybe once or twice per year. I do it online or in an app, I hate the process. But automatic bill payment takes care of 99% of my bills!
Arent these basically minimum wage jobs? I mean throw a few dollars an hour on top of that, but there are plenty of jobs like this.
Any time I needed anything advanced, I get shuffled to someone else.
What is a Bank nowadays. It is nothing. It is a virtual construct and software that we are supposed to put our trust into, where banks have a history of betraying that trust.
I don't feel the phone conclusion is quite correct, because it's not just the need to use an ATM that has dropped. The need to use a banking app or website has also dropped.
The behavior of companies has changed dramatically. Checks have almost vanished, you can often set up automatic payments, and you can get bank balance notification emails/messages. A large portion of banking interactions are fully automated.
Am I weird in that I don't think I ever interacted with a bank teller?
The only bank employee I ever interacted with was when getting a mortgage and maybe opening an account, but at least here in the Netherlands, I don't think there are any bank tellers left really? Old people complain of course, but am I missing something here?
EDIT: I'm pushing 40, relevant here I guess
I hate the graph here. "Bank teller employment has fallen off a cliff" - well it _looks_ that way but actually it's more like halved from its peak because the bottom of the Y axis isn't zero. That's still a significant reduction, but it's not as dramatic as it seems at first glance.
Lies, damn lies...
Misleading graph alert! The green graph's vertical axis starts at 150,000 instead of 0. It shows the number falling to about 50%, not 10% as it appear at first. The misleadingness fits the author's narrative, which is how it always seems to go so I think it's safe to assume malice and that he's trying to mislead his readers.
The graph showing that "Bank teller employment has fallen off a cliff" is not zero based. This is pretty damn bad. The graph looks like it's going down 90%, but it's actually going from 350k to 150k. That's a ~60% drop which is a lot, but not "falling off a cliff".
If I have to physically still go to the bank, it really hasn't disrupted much. The iPhone created an opportunity... the banks investing around the technology is the disruption. ATM itself couldn't unlock as much which I suppose is the paradigm mentioned in the article.
AI is more iPhone than ATM IMO.
As AI gets better the bottlenecks will be the place to watch. Bottleneck jobs will become more productive => they either pay more or more bottleneck jobs will be created or some in between situation occurs. This will continue until no bottlenecks are left.
My experience of building automations with AI is that we aren't yet at the ATM phase. AI is going to help us to automate existing processes, but it's not the technology that will displace humans.
That technology doesn't exist yet.
*Mobile phones. Calling and texting your bank for account info and actions predates the iPhone... even Venmo started as a text message service before it was an app. The iPhone may have just been the nail in the coffin.
This writing style where every section has multiple paragraphs of preamble, prolepsis, cold openers for cold openers, and tangents is infuriating. Get on to the point already.
But there are still people being paid with the job title bank teller. slowed growth would be more accurate. Maybe ATMs also slowed would-be growth? Impossible to say.
The author wants to say that atms are a stand in for in person banking experience, while the iPhone changes the paradigm entirely.
Why? Seems like basically the same paradigm to me, I can just do it without going anywhere.
I guess the trope in movies of masked bank robbers going in and threatening a scared bank teller will be a thing of the past soon. Pointing a gun at an iPhone doesn't have the same vibe.
I really enjoyed this article, I didn't bridge the idea of an ATM and mobile banking.
I think the idea raised about "Automated Firms" is a bit off in the picture painted in that linked article. I think the David Oks intention is to paint a picture of a fully automated company, but the linked article gives this impression:
> Future AI firms won’t be constrained by what's scarce or abundant in human skill distributions – they can optimize for whatever abilities are most valuable. Want Jeff Dean-level engineering talent? Cool: once you’ve got one, the marginal copy costs pennies. Need a thousand world-class researchers? Just spin them up. The limiting factor isn't finding or training rare talent – it's just compute.
In that above paragraph the author is saying to the reader that a human will be able to spin up and get these armies of intelligent workers, but at the end of the day their output is given to a human who presumably needs to take ownership of the result. Intelligent workers make bad choices or bad bets, but those AI machines cannot "own" an outcome. The responsibility must fall on a person.
To this end, I think the fully autonomous firm is kind of a fallacy. There needs to be someone who can be sued if anything goes wrong. You're not suing the AI.