Which Jobs Will Still Exist in 2040 — My Honest Guess

“Take the bank one,” I said.

My cousin went quiet on the other end of the call. She’d asked me something I had no business answering that confidently: which of the two jobs in front of her would still exist in 2040?

It was a Sunday in March. Option one was an operations analyst role at a regional bank. It paid $58,000 with health insurance, and the manager had already said “long-term fit.” Option two was a spot in a 16-month accelerated nursing program that would cost around $38,000 she didn’t have. Her parents had voted for the bank, and so had her friends. I was the last vote, the relative who writes about the future on the internet. I voted for the bank too.

(I’ve blended details from three conversations I had this year into one. The numbers below are real, and so is what I said.)

I haven’t stopped thinking about that call. My information wasn’t wrong. The problem was the framing, which hid a cost, and I suspect most people make career decisions inside that same framing.

Which jobs will still exist in 2040?

Most jobs that need a human body in the room, a professional license, or trust between strangers will still exist in 2040. Think nursing, the skilled trades, therapy, early education and energy installation. Plenty of office jobs will survive as well. There will just be far fewer ways into them.

The official projections aren’t subtle about this. The U.S. Bureau of Labor Statistics’ 2025–35 projections rank nurse practitioners as the fastest-growing occupation in the country, at 41%. Solar photovoltaic installers come next at 37%, then data scientists at 35%, wind turbine service technicians at 30% and physical therapist assistants at 23%. As of May 2025, BLS put the median pay for nurse practitioners at $132,300.

The World Economic Forum’s Future of Jobs Report 2025 surveyed more than 1,000 employers with over 14 million workers between them. They expect growth in care roles, teaching, delivery driving and farm work. Clerical work tops the list of declining roles: data entry clerks, bank tellers, administrative assistants.

So on paper the nursing program was the better bet, and I knew that when I told her to take the bank job. I said it anyway because the bank job was legible. It came with a salary, a title and a manager who liked her. The nursing program came with a tuition bill and 16 months of no income. One of them looked like risk and the other looked like stability.

I never asked what the bank job actually involved.

Is AI already taking jobs from young workers?

Across the whole economy, not yet. At the entry level it already is, and it’s happening through hires that never get made rather than through layoffs.

The clearest evidence comes from the Stanford Digital Economy Lab’s “Canaries in the Coal Mine” research by Erik Brynjolfsson, Bharat Chandar and Ruyu Chen. Their August 2026 update uses ADP payroll records for millions of U.S. workers through June 2026. In AI-exposed occupations, employment of 22- to 25-year-olds is now 19% below where it would be if it had kept pace with their less-exposed peers. Experienced workers in the same occupations show no comparable gap. The gap has kept widening since the team first reported it in August 2025.

How it happens matters more than the headline number. The authors found the gap “operates primarily through reduced hiring of young workers rather than increased separations.” Nobody gets fired and nothing makes the news. Junior postings just quietly stop showing up.

I know this from the employer’s side because I’m part of it. I run three products by myself. Five years ago that workload would have meant a part-time support hire and a junior developer at minimum. I hired neither. Nobody lost a job because of me, but two jobs that would have existed were never posted. Multiply that by a few hundred thousand small companies and you get a 19% gap without a single layoff announcement.

Then I reread my cousin’s offer letter. An operations analyst at a regional bank spends the day on reconciliations, exception reports, document checks and moving data from one system into another. You’d struggle to write a tidier list of the tasks being automated first.

What is the hidden cost of choosing the safe career?

You pay for the safe career later, usually in your early thirties. The role gets restructured and you end up job-hunting with skills priced for work that has stopped hiring.

Economists have a decent idea of how big that bill is. In “Recessions and the Costs of Job Loss” (Brookings, 2011), Steven Davis and Till von Wachter followed men who lost jobs in mass layoffs. The ones laid off while national unemployment was below 6% lost an average of 1.4 years of pre-layoff earnings over the next 20 years, in present-value terms. The ones laid off while unemployment was above 8% lost 2.8 years.

Now run my cousin’s numbers. On a $58,000 salary, 1.4 years of earnings comes to $81,200. In a weak economy, 2.8 years comes to $162,400. The nursing program cost $38,000. So the “risky” option cost less than half the expected loss from the “safe” one, and I haven’t even counted the pay gap between the two careers.

Skill decay makes it worse. The WEF expects 39% of workers’ core skills to change by 2030. Say you spend ages 23 to 28 doing reconciliation work that software is learning to do. You come out with five years of experience in a shrinking category, and those were the years when your earnings should have been compounding fastest.

Both options send a bill. The risky one sends it up front. The safe one sends it around 34, with interest. We’re bad at comparing the two because only one of them is printed on a tuition page. No offer letter has a line for displacement risk.

Hasn’t technology always created more jobs than it destroyed?

For the economy as a whole, yes. That doesn’t help any one person much, because the new jobs mostly go to different people than the ones who lost the old jobs.

This is the best counterargument, and the evidence for it is serious. In “New Frontiers” (NBER 2022, Quarterly Journal of Economics 2024), David Autor and his co-authors found that about 60% of U.S. employment in 2018 was in job titles that didn’t exist in 1940. The WEF projects 170 million jobs created and 92 million displaced by 2030, a net gain of 78 million. The Yale Budget Lab found no discernible disruption to the broader labor market in the 33 months after ChatGPT launched. Its measures of AI exposure showed no link to changes in employment or unemployment.

I believe all of it, and it still doesn’t answer my cousin’s question.

It helps to look at what each study measures. Yale looked at the occupational mix and unemployment rates of people already in the workforce. A 23-year-old who never got hired doesn’t register as displaced. She just turns up somewhere else, in a different job or another degree. Using the same payroll data, the Stanford team found no economy-wide displacement and a 19% gap for young workers. Both findings hold at once. The overall numbers look stable because the cost lands on people who haven’t entered the workforce yet.

The WEF’s net 78 million has the same blind spot. The 92 million people who get displaced aren’t the 170 million who get hired. Davis and von Wachter’s workers lost 1.4 years of earnings during periods when the overall economy was adding jobs. When desktop publishing spread, the new jobs went mostly to young people just starting out. The typesetters didn’t get them.

“Technology creates more jobs than it destroys” describes economies. It was never a promise to any particular worker.

Which careers are safest for young people starting out in 2026?

The safest careers are the ones where AI makes experienced workers more valuable and juniors are still needed. That’s usually because the work needs a person on site, a license, or someone who is legally accountable.

Here’s my honest guess for 2040. I’ve sorted jobs by whether they’ll still take in new people, because a job can exist and still stop hiring.

These will exist and still hire at the bottom: nurse practitioners and registered nurses, electricians, grid and battery technicians, physical and occupational therapists, mental health counselors, early childhood teachers, and anyone whose signature carries legal liability for work a machine drafted. A machine can take on more of the task, but somebody still has to answer for it.

These will exist and pay well, but they’ll be much harder to get into: software engineering, law, accounting and financial analysis. Data scientists sit near the top of the BLS growth list, so office work isn’t going anywhere. What’s changing is who gets hired. By 2040 I expect these fields to recruit mostly experienced people, and getting in will look more like an apprenticeship than a graduate job.

These will shrink hardest: data entry, tier-one customer support, back-office reconciliation and document processing. That’s pretty much the job my cousin was about to start.

Then there’s the category nobody can list. If Autor’s pattern holds, a big share of 2040’s jobs don’t have names yet. You can’t train for those directly. Your best bet is to work somewhere that gives you real responsibility early.

The question I ask now isn’t “will this job exist in 2040?” It’s “will this job still need a 25-year-old in 2030?” Plenty of jobs will make it to 2040 without ever hiring you.

Should you take the safe job or retrain for a growing field?

If the safe job is made of the tasks being automated first, taking it is the bigger risk. Retraining for a licensed, in-person field costs more now and a lot less over a career.

She took the bank job. Six months in, two analysts on her team left and neither was replaced. The reconciliation work moved to a vendor tool. She now spends her afternoons checking what the tool flagged and teaching it edge cases. Last week she called the job “fine, just quieter.” Then she asked whether the nursing program still took spring applicants.

I can see now what I did that Sunday in March. I thought I was steering her away from risk. What I actually did was pick the risk with the invisible bill, since the other option’s bill was posted on a website. “Take the bank one” felt like caution. Really, it was the option nobody had put a price on.

This time I sent her the application page. The deadline is November 1.