Jobs that will be replaced by AI in 2026: what five major studies say

If you had to bet on one job shrinking because of AI, data entry keyer would be a safe pick. Going by FinanceBuzz's read of Bureau of Labor Statistics projections, the role is set to lose nearly 26% of its jobs by 2034, and Anthropic's March 2026 labor study found Claude already covering 67% of its tasks. The wider answer follows the same pattern: the jobs that will be replaced by AI first are routine, screen-based ones, and five of the most-cited studies largely agree on which they are.

They don't agree on how fast, though. And they measure very different things, which is why I've lined them up role by role below instead of picking one headline number.

Short answer: which jobs will be replaced by AI?

The jobs most at risk from AI are customer service representatives, data entry keyers, administrative and legal assistants, bookkeeping clerks, cashiers, proofreaders, telemarketers, credit analysts and routine programming roles. Hands-on and high-trust work (cooks, mechanics, pharmacists, clergy, air traffic controllers) is the safest. Most studies expect tasks to disappear well before whole occupations do.

So "replaced" usually means fewer people hired into a role, not everyone fired at once.

The background: five studies asking five different questions

It helps to know what each study actually counts, because they're often quoted as if they were the same kind of evidence. They aren't.

The World Economic Forum's Future of Jobs Report 2025 is a survey of employers. It estimates 170 million jobs created and 92 million displaced worldwide by 2030, a net gain of 78 million. AI and information processing alone are expected to create 11 million jobs and displace 9 million. The fastest-declining roles in its survey include cashiers and ticket clerks, administrative assistants and executive secretaries, printing workers, and accountants and auditors.

The IMF looked at exposure across whole economies. In January 2024, managing director Kristalina Georgieva wrote that "almost 40 percent of global employment is exposed to AI", rising to about 60% in advanced economies. Roughly half of those exposed jobs may benefit from AI; for the other half, she warned, lower demand could mean lower wages and reduced hiring, and "in the most extreme cases, some of these jobs may disappear."

Goldman Sachs Research put a number on displacement. Economists Joseph Briggs and Sarah Dong estimate that AI could displace 6-7% of the US workforce if it's widely adopted, with a range of 3% to 14% under different assumptions. They expect the effect to be temporary, lifting unemployment by about half a percentage point during the transition. (They also point out that about 60% of US workers today are in occupations that didn't exist in 1940, which is a useful number to keep in your back pocket.)

Anthropic's Economic Index uses real usage data from Claude rather than forecasts. Its March 2026 report found that about 49% of jobs have seen at least a quarter of their tasks performed using Claude. Anthropic makes Claude, so it's measuring its own product; that's worth keeping in mind, even if the data is unusually direct.

Finally, the BLS builds official US projections. For 2024 to 2034 it projects total employment growing 3.1%, to 175.2 million, while office and administrative support occupations shrink 3.9%, a loss of 761,900 jobs. The agency names generative AI as one force expected to dampen demand in sales, design and admin support.

Exposure scores work a bit like a flood map. They show which houses sit low, not which ones will flood this year, and whether yours does depends on things the map can't see.

Real examples: the jobs most likely to be replaced by AI, study by study

Here's our own cross-reference of the five sources. A blank cell doesn't mean a study rates the job safe; often the role simply isn't broken out.

Sources: WEF, Goldman Sachs, Anthropic, BLS and FinanceBuzz's summary of BLS data. Cashier losses owe a lot to self-checkout, not just AI.

Customer service is the one role every usage-based source flags. Anthropic says the main tasks of customer service representatives increasingly show up in its API traffic, and a Built In write-up of the latest Economic Index notes that API workflows for business sales outreach and automated trading doubled between November 2025 and February 2026. If that keeps up, sales reps, traders and market researchers could be the next group on these lists.

The disagreements are more revealing than the overlaps. Computer programmers top Anthropic's exposure ranking, yet BLS expects software developers to add 267,700 jobs. Accountants appear on the WEF's declining list, while BLS sees accountants growing and bookkeeping clerks shrinking. My read is that the routine, junior version of a job shrinks while the judgement-heavy version grows. That fits the entry-level squeeze we covered in October, and Goldman's finding that unemployment among 20- to 30-year-olds in tech-exposed occupations has risen by almost 3 percentage points since the start of 2025. If you're an engineer weighing this, our piece on whether engineers will be replaced by AI goes further.

At the other end, Goldman lists air traffic controllers, chief executives, radiologists, pharmacists, residential advisors, photographers and clergy as the least likely to be displaced. Anthropic found that 30% of workers have zero coverage in its data, including cooks, motorcycle mechanics, lifeguards, bartenders and dishwashers. Robots aren't closing that gap quickly either: Anthropic's robotics index says robots can already do three-quarters of physical tasks, mostly in limited settings, but are cost-competitive for just 0.3% of them. For the full list of safer roles, see jobs AI can't replace and jobs that won't be replaced by AI.

Benefits and limitations of the forecasts

The strongest point in their favour is agreement. When an employer survey, a usage log, an investment bank's model and a government projection all point at customer service and admin work, I'm fairly confident that's real. Anthropic even checked its measure against BLS: for every 10 percentage point rise in task coverage, the BLS growth projection drops by 0.6 points.

That relationship is slight, and Anthropic says so itself. The other limits are just as plain. The WEF numbers are what executives expect, not what has happened. The IMF's 40% includes jobs that may gain from AI. Goldman found no significant statistical link so far between AI exposure and job growth, unemployment or layoff rates, partly because only 9.3% of US companies in one survey had used generative AI in production in the previous two weeks. And Anthropic sees only Claude.

Real-world layoffs are smaller than the exposure numbers too. US employers blamed AI for 116,175 announced job cuts between January and August 2026, according to the Challenger, Gray & Christmas August report. That's a lot of people. It's also a small share of a 170-million-job economy, and our guide to what jobs AI will replace looks at why some of those announcements may be AI-washing.

Practical takeaways for your career

Start with your tasks, not your job title. Write down what fills your week. If most of it is reading documents and re-entering data, answering the same questions, or drafting routine text, you're in the zone every study flags; if it's physical, relational or full of judgement calls, you're probably further from it than the headlines suggest.

Next, move toward the judgement end of your own field. The BLS split between bookkeeping clerks and accountants, or between programmers and software developers, shows where the growth sits. Security is another growth pocket: BLS projects information security analysts up 28.5%, which ties into our look at whether AI will take over cyber security jobs.

Get good at the tools. Anthropic found people who'd used Claude for six months or more had a 10% higher success rate in their conversations, even after accounting for task and country.

And honestly, I think hands-on licensed trades are underrated by people choosing careers right now. They sit at the bottom of nearly every exposure list, and the robotics data suggests that won't change soon.

The next two years

I'd expect the clearest change to happen inside occupations rather than between them. Anthropic reports that coding work is moving from chat in Claude.ai to automated API workflows, and that kind of shift probably reaches customer support and sales operations next. Goldman also warns that if adoption is more frontloaded than its economists assume, the unemployment bump could be bigger than half a point.

What I'm watching is hiring, not layoffs. The first sign of replacement in the data has been fewer junior openings in exposed roles, and that's the number I'd track in your own field. Briggs and Dong say which jobs get displaced "will remain an open question" until the AI adoption cycle has fully played out. That seems right to me.

FAQ

What types of jobs will AI affect the most?

Office and administrative support work is hit hardest across the studies. That includes customer service, data entry, secretarial work, bookkeeping and routine programming. BLS projects office and admin support jobs falling 3.9% by 2034, and the IMF estimates about 60% of jobs in advanced economies are exposed in some way, though roughly half may benefit.

What jobs can AI not replace?

No study calls any job fully safe. Physical and high-trust roles rank lowest, though: Goldman Sachs names air traffic controllers, radiologists, pharmacists and clergy, and Anthropic's data shows zero coverage for cooks, mechanics and bartenders.

What jobs will AI take over?

AI is most likely to take over tasks inside customer service, data entry and admin roles first. Anthropic finds Claude covering 75% of computer programmer tasks and 67% of data entry tasks. Taking over tasks isn't the same as taking over the job, though: BLS still projects software developer jobs growing 15.8% to 2034.

Which jobs will be replaced by AI?

The roles with the clearest projected declines are data entry keyers (about 26% by 2034), cashiers (about 10%), legal secretaries and admin assistants (5.8%) and customer service representatives (5.5%), going by BLS projections. Goldman Sachs adds proofreaders, telemarketers and credit analysts to its high-risk group. Most of these losses are expected to come through slower hiring and roles that aren't refilled when someone leaves, so people already in these jobs probably have more time than the headlines imply.

Which jobs are most at risk from AI?

Customer service representatives are the one role flagged by almost every source: Goldman rates them high risk, Anthropic ranks them second for exposure, and BLS projects a loss of 153,700 jobs by 2034. Computer programmers, administrative assistants and accounting clerks follow close behind on most lists.