The search for an AI-proof job
Young people are right to be worried about the market these days
Young Americans are much more worried than most that artificial intelligence could take their jobs. And to be blunt, they are absolutely right to be.
For all the apocalyptic predictions that chatbots could soon wipe out vast swaths of employment, the majority of U.S. adults aren’t exactly panicked yet. In The Argument’s latest monthly poll, just 17% of registered voters thought AI could replace them at work at this moment. The share increased to 24% when asked about five years from now. If a quarter of all jobs were actually taken over by AI, it would, of course, be an economic earthquake. But as a measure of public nervousness, that’s pretty moderate.
But participants’ confidence in their job security decreased with their age. Among those under the age of 30, 26% think their work could be handled by AI today, and 29% think it will be possible within half a decade. The teens and twentysomethings were also much more likely than other groups to admit they were unsure whether AI could steal their spot at the office, suggesting a general sense of wariness about where the future is headed.
Familiarity breeds anxiety here. As my colleague Lakshya Jain wrote last week, young adults are simply much more likely to use AI on a day-to-day basis, and thus have a deeper sense of its capabilities. The idea that one marketing staffer could soon do the work of six may seem more self-evident to people who spent their college years writing papers with the help of ChatGPT. (Especially if they themselves came to rely on it as a crutch.)
But it’s also possible that young people are just accurately reading the state of the labor market. There’s little to no evidence that the rise of large language models has led to any kind of broad-based increase in unemployment, even for occupations where the workers could, in theory, be most easily replaced by an AI interface. But there are signs it may be creating a crunch for entry-level workers.
Last month, a working paper by economists at Stanford showed that, since LLMs arrived on the scene, employment among entry-level workers has been stagnant or dropping in the occupations most exposed to AI-driven automation. Yet job growth continued apace in other industries and for older workers.
Take coding. The authors found that between late 2022 and July 2025, “employment for software developers aged 22-25 declined by nearly 20%,” even as it rose for workers aged 30 and older. They identified a similar trend in customer service, where audio chatbots have quickly started to gain traction, as well as marketing and sales managers, who do a lot of the white-collar work that AI is able to automate.
In more physical occupations, like health aides and production workers, jobs grew across age groups — another sign that AI, rather than a broader macroeconomic trend, may be the issue in fields like software and marketing. But individual pockets of trouble add up: The Stanford team found that altogether, the headcount of early-career workers has been “stagnant.”
Why might AI be killing jobs for younger workers but not older ones? It may be that these models are excellent at automating white-collar grunt work but can’t replace the more sophisticated decision-making skills of more experienced managers. But it’s also possible that as companies get used to the technology, executives will feel more comfortable replacing other parts of their workforces, in which case young people will just be the first domino to fall.
If the Stanford findings are correct and AI does turn out to be winnowing entry-level opportunities, what should we do about it? There are not a lot of easy answers, in part because we still have little sense of how AI will really ripple through the broader economy.
Maybe Anthropic CEO Dario Amodei isn’t just talking his book and half of all entry-level white-collar jobs really are at risk of being digitized. Maybe Elon Musk’s plan to replace the world’s factory workers with AI-enabled robots will come true and self-driving technology will make Uber drivers a job of the past.
RELATED: Please let the robots have this one
Or maybe the bottom rung of corporate America is just going to get slightly smaller, making competition for jobs at Google and Meta fiercer. We’re sort of in the realm of sci-fi turning real and it’s hard to plan.
My own guess is that LLMs will probably change how we do office jobs but won’t lead to the end of white-collar work as we know it. But if I were giving advice to a high schooler who was genuinely worried about AI and trying to plan their future around it, here’s what I would say:
First, if you’re on the fence about whether college is even the right path for you, keep in mind that ChatGPT won’t be responding to any calls for a plumber or electrician anytime soon.1 Learning to work with your hands — and no, typing doesn’t count — is probably a better hedge than knowledge work since our advances in LLMs have far outpaced our advances in robotics. If you would be happy in a skilled trade (emphasis on if, since these jobs are hard and definitely not for everyone), it will probably be a great option for a long time.
Second, consider working in health care.
Health care jobs — with their combination of cognitive work and high-touch patient interactions — are expected to be fairly resistant to automation. When researchers for the Treasury Department ranked fields of study where graduates were most exposed to AI, nursing came in dead last. Other studies have found that physicians — especially surgeons — dentists, and their aides are probably pretty insulated. Occupational and physical therapists also were fairly safe.
Even health care occupations that were widely expected to be endangered thanks to AI models have turned out to be bot resistant. In 2016, the computer scientist Geoffrey Hinton warned that medical schools should “stop training radiologists now” because computer image recognition was becoming so accurate and powerful. But as Deena Mousa recently wrote at Works in Progress, “demand for human labor is higher than ever” in the field, with residency programs expanding and average salaries still topping $500,000.
The reasons vary: While AI models are excellent at detecting tumors and other pathologies in controlled settings, they are less successful in the field. They also can’t handle the task of working with other doctors. Plus, insurers are leery of paying for 100% bot-based care. Probably work on your bedside manner though; one of the value adds of having a human doctor is, well, that they express human emotions.
Even if AI turns out to be a bust, these hedges are just good life advice. America is going to need a lot of health care workers and electricians. Medicine is already one of the few bright spots in the economy keeping job growth afloat. And unless the deep state really has been hiding the medbeds from us all, aging baby boomers are going to need more care in the future. As a result, we’re facing a shortage of professionals like primary care doctors, nurses, and electricians. These are not easy jobs — nursing in particular is notorious for burnout — but obviously lots of people find them immensely rewarding.
Finally, some more general advice, as The Argument columnist Derek Thompson has argued: avoid letting AI de-skill you. Letting your classmate do all the work on a project can feel smart in the short run, but in the long run, the joke’s on you. The same goes for an LLM. So do your biology homework — and not with ChatGPT.
"…avoid letting AI de-skill you."
This is great advice. I submit that the same could be said for the internet, generally — including and especially the mobile internet.
Someone once said that human beings became cyborgs the moment our ancestors picked up their first tool, because every tool is simply a cybernetic extension of our bodies. As humans have grown more cybernetically enhanced with more sophisticated tools — including tools which, in turn, are designed to direct other tools (as computer operating systems direct the various software programs which use them) — we have become relatively less human, and more machine. The trick is to let our tools help us with the work we do, and not to largely do this work themselves — or, rather, do it with minimal input.
If you've ever noticed someone who seems hopelessly lost without their cell phone — or, perhaps if you've ever felt lost yourself under this circumstance — then you know just what I mean…
Most of the things that AI seems to be good at, at least with LLM, is the dumb grunt work that few people wanted to do and made even less sense economically. Which means maybe that work wasn’t worth much to begin with. For example, my wife has used it to compose work emails that she hates writing. I’ve used to compose other work correspondence that’s required but that nobody reads.
I’ll go even further and say that maybe it’s revealed how shallow and not worthwhile some of work has become. For example, if I can feed a hundred Atlantic columns into a machine and then it spits me out a new one that I can’t tell the difference from, maybe those essays weren’t really all that important or thought provoking.
I don’t think AI is going to replace humans in the important ways. But I do think it’s going to reveal that we haven’t really been working or thinking in important ways in some jobs.