Jobs AI will replace and jobs it can't: what the data shows

By Johannes Sundlo, AI and future of work advisor, formerly HR at Spotify. Updated 2 November 2026.

The short answer

AI takes over tasks more often than whole jobs. The most exposed jobs run on text, code, numbers and conversations that follow a pattern, like data entry, customer service, translation, writing and programming. In US projections to 2035, word processors and typists, telephone operators, data entry keyers and telemarketers are among the fastest-shrinking occupations. The least exposed jobs combine physical work, hands-on care and responsibility for people. Employment data shows no broad job losses yet, even though AI was the most-cited reason in US layoff announcements so far in 2026. My advice is to look at your tasks before your job title.

Check your own job

Most exposed Office support, customer service, writing, translation, programming.
Least exposed Hands-on healthcare, construction, repair, cooking, cleaning.
Already visible Fewer entry-level hires in the most exposed jobs.

Which jobs will AI replace?

The jobs most likely to shrink are office and customer-contact jobs where AI can do much of the work. In the US Bureau of Labor Statistics (BLS) projections to 2035, word processors and typists (−34.4%), telephone operators (−27.6%), data entry keyers (−25.5%) and telemarketers (−21.4%) are among the fastest-declining occupations. Office and administrative support is the fastest-shrinking major group, down 4.0%, and the BLS expects automation, including AI-based tools, to reduce demand for several of its occupations.

Researchers measure exposure in different ways, from what AI could do in theory (ILO, Goldman Sachs) to what people do with Bing Copilot (Microsoft) and Claude (Anthropic), and still point at many of the same jobs.

Exposure	Type of work	Examples	Why
High	Office and administrative support	Data entry keyers, typists, file clerks, payroll clerks	Rule-based work with text and numbers. The most exposed group in the ILO index, and data entry keyers are third on Anthropic’s list.
High	Customer service and sales	Customer service reps, telemarketers, sales reps for services	Conversations that follow a pattern. Customer service reps are second on Anthropic’s list, and all three are in Microsoft’s top 10.
High	Writing, translation and editing	Interpreters and translators, writers, editors, technical writers	AI writes, edits and translates. Interpreters and translators top Microsoft’s list.
High	Programming and data	Computer programmers, web developers, data scientists	Computer programmers top Anthropic’s list, with 75% coverage in its observed-exposure measure.
High	Legal, finance and analysis	Legal staff, management analysts, financial advisors, tax examiners	Documents, rules and analysis. Goldman Sachs estimates that 44% of tasks in legal jobs could be automated, and management analysts and financial advisors are in Microsoft’s top 40.
Mixed	Teaching	Schoolteachers, college teachers	Several kinds of college teachers are in Microsoft’s top 40. But as Anthropic puts it, AI “can grade homework but not manage a classroom.”
Low	Hands-on healthcare	Phlebotomists, surgical assistants, orderlies, massage therapists	Physical presence, licenses and patients. All four are in Microsoft’s bottom 40. AI helps with the notes.
Low	Construction, installation and repair	Roofers, cement masons, tire repairers, highway maintenance workers	Physical work in places that keep changing. Goldman Sachs puts exposure at 6% of tasks in construction and 4% in maintenance.
Low	Food, cleaning and hospitality	Cooks, bartenders, dishwashers, housekeeping cleaners	Cooks, bartenders and dishwashers had zero coverage in Anthropic’s data.
Low	Operating vehicles and machinery	Forklift operators, logging equipment operators, motorboat operators	Physical control of a machine. All three are in Microsoft’s bottom 40.

Sources: ILO (May 2025), Microsoft Research (Bing Copilot conversations from January to September 2024), Anthropic (March 2026), Goldman Sachs (March 2023) and BLS projections (August 2026).

Which jobs can’t AI replace?

There are no AI-proof jobs, but jobs that combine several of these traits are the hardest to replace.

Physical work in places that change, like roofers, home health aides or wind turbine technicians.
Licenses and legal accountability, where society wants a named person to own the decision, like nurse practitioners or surgeons.
Trust and relationships, like caring for someone, teaching, leading or having the difficult conversation.
Judgment when the situation is unclear, where there’s no right answer to calculate, like a fire officer deciding how to attack a fire or a manager handling a conflict in the team.

The lowest scores in the research go to jobs like phlebotomists, orderlies, massage therapists, roofers, cement masons, dishwashers and logging equipment operators in Microsoft’s data, and cooks and bartenders in Anthropic’s. Among growing jobs, nurse practitioners, wind turbine service technicians and home health aides combine several of the traits above. AI helps in these jobs too, with the notes and the quotes, but the patient and the roof remain the person’s responsibility.

Will AI take my job?

For most people, AI will take parts of the job. Exposed means AI can do or speed up part of the work, and it says nothing yet about what happens to the job. That depends on whether demand grows when the work gets cheaper, on what your employer does with the freed-up time, and on forces that have nothing to do with AI. The BLS, which now rates 831 occupations from low to very high AI exposure, states that “exposure does not imply job loss.” The projections show all four combinations.

AI exposure	Projected to shrink by 2035	Projected to grow by 2035
High	Telephone operators −27.6%, data entry keyers −25.5%, telemarketers −21.4%. AI can do much of the work, and employment is projected to fall. Here the risk of replacement is highest.	Data scientists +34.6%, information security analysts +21.0%. The work changes, but demand grows.
Low	Foundry mold and coremakers −23.0%. Microsoft scores the job at zero for AI, so other forces drive the decline.	Ophthalmic medical technicians +21.4%, massage therapists +15.3%. Hands-on work in fields that are growing.

Projections from the BLS, 2025–35 (August 2026). Exposure from Microsoft’s top and bottom 40, and from Anthropic’s top ten for data entry keyers and information security analysts.

Anthropic compared its own measure with the BLS’s earlier projections to 2034 and found only a slight link, with 0.6 percentage points less projected growth for every 10 points of higher coverage. So I’d read high exposure as a sign that your job will change, and the projections as a sign of whether there will be as many jobs like it. To place your own job, check its rating in the BLS AI exposure categories and its projected growth in the Occupational Outlook Handbook, then run the check below on your own tasks.

If the question worries you personally, start with my notes on what to do when the AI anxiety is your own.

What does the future of jobs look like?

The BLS projects US employment to grow 3.5% from 2025 to 2035. Healthcare makes up more than half of the 30 fastest-growing occupations, and clean energy, data and security hold several of the top places.

Occupation	Growth 2025–35	AI can help with	Stays with the person
Nurse practitioners	+41.0%	Notes and patient letters	Examining patients and deciding on treatment
Solar photovoltaic installers	+36.5%	Quotes, layouts and permit paperwork	The installation on the roof
Data scientists	+34.6%	Code, first analyses and documentation	Choosing the question and owning the conclusion
Wind turbine service technicians	+29.5%	Fault finding and reports	The climb, the repair and the safety calls
Physical therapist assistants	+23.0%	Documentation and exercise programs	The hands-on treatment
Information security analysts	+21.0%	Log reviews and summaries	Judging the threat and deciding what to do
Home health and personal care aides	+18.1%	Scheduling and notes	The care
Industrial machinery mechanics	+17.8%	Fault finding and manuals	The repair

A selection from the 30 fastest-growing occupations in the BLS projections (August 2026). The two AI columns are my assessment.

For readers in England, National Careers Service job profiles show day-to-day tasks, salary and routes in.

Check how exposed your own job is to AI

Do the math on your own working time. Two people with the same title can get very different answers.

List your tasks. Write down 10 to 15 things you do in a normal month and roughly how many hours each takes. If you get stuck, O*NET OnLine lists typical tasks for more than 900 occupations.
Test them for real. Ask the AI tool your employer has approved (anything else is shadow AI) for a first draft of each task, with made-up names and details for anything about people. Mark yes if you could use the draft after a read-through and a few edits, partly if it’s a good start but you rewrite large parts or it lacks information only you have, and no if the task needs you on site, making the call or meeting a person.
Work out the share of your time. Count yes tasks in full and partly tasks at half. Under 10% means your job changes a little, 10 to 49% that AI becomes a tool in your job, and 50% or more that the job is likely to change at its core. I borrow the 10 and 50% cut-offs from the 2023 Goldman Sachs report behind the often-quoted 300 million jobs, which used them for illustration and labeled jobs at 50% or more as likely to be substituted. I read the top band more softly, because you’re measuring your own hours, not a whole occupation.
Look at what’s left. The tasks you marked no or partly are where your job is heading. Get better at them, and make sure people see you doing them.
Check demand, and repeat. Look up your occupation in the Occupational Outlook Handbook or, in England, with the National Careers Service. Do the exercise again in six months, when the tools have improved.

This prompt works for each task in step 2. The more context you give, the fairer the test, and my guide on how to write a good prompt shows what to add.

I work as a [job]. One task I do every month is [describe the task, who it’s for and what the result should be]. Write a first draft. Then tell me what information or material you’d need to do it well.

Here is a made-up example, a marketing coordinator who spends 100 hours a month on six tasks.

Task	Hours per month	Good first draft from AI?
Write social posts, newsletters and web copy	25	Yes
Pull campaign numbers into a monthly report	10	Yes
Plan events with venues and suppliers	20	Partly
Brief agencies and review their work	15	Partly
Meet sales and customers to find stories	20	No
Staff the booth at trade shows	10	No

That’s 35 hours of yes and 35 hours of partly. With partly counted at half, the coordinator lands at 52.5%, just over the 50% line. The hours will move toward events, customers and agencies, so that’s where the coordinator should build skills.

Your result	What it means	Do this
Under 10%	Your job changes a little	Learn the tool for the tasks where it helps, and repeat the exercise in six months.
10–49%	AI becomes a tool in your job	Agree with your manager what the freed-up time goes to, so it doesn’t fill up with more of the same.
50% or more	Your job is likely to change at its core	Talk to your manager about what the role looks like a year from now, pick the remaining tasks you’ll get best at, and ask what training or transition support you can get.

What has happened to jobs so far?

AI shows up in layoff announcements, but it barely shows in overall employment.

Layoff announcements. US employers cited AI as the reason for 120,136 announced job cuts from January to September 2026, about 21% of all announced cuts and the most-cited reason so far this year, according to the outplacement firm Challenger, Gray & Christmas. That’s less than a tenth of a percent of the roughly 170 million jobs in the US. An announcement is the company’s stated reason for a plan, not a measured effect of AI.
Across the OECD. The OECD’s June 2026 Economic Outlook found no signs of widespread job displacement from businesses adopting AI at the industry level. In most countries with data, vacancies grew more in the most AI-exposed industries than in other sectors over the year to April 2026, with the US as a notable exception.
In the US. Stanford’s Digital Economy Lab, using ADP payroll data on millions of workers through June 2026, finds no evidence of widespread job displacement. Anthropic found no systematic rise in unemployment for the most exposed workers since late 2022.

The studies cover the few years since ChatGPT launched in late 2022, so anyone who tells you for sure what happens in five years is guessing.

What is happening to entry-level jobs?

The clearest change so far is at the entry level. In the Stanford data, employment for 22–25-year-olds in AI-exposed jobs is 19% below where it would be if it had kept pace with less exposed peers, and the gap has widened since it was first documented in August 2025. It comes mainly from fewer young people being hired. Experienced workers show no such gap, and where AI mostly complements workers, employment is flat or rising. The authors call their findings early, descriptive indicators and not causal estimates. The gap shrinks when they control for education, some of the divergence predates generative AI, and it’s larger in the ADP data than in national surveys. Anthropic also sees suggestive evidence that hiring of younger workers has slowed in exposed jobs.

Swedish job ads point the same way. In my report Vad händer med de juniora jobben (in Swedish, August 2026), all ads for graduate-level jobs fell about as much from 2022 to 2025 in AI-exposed occupations like IT, finance, legal and HR (−39%) as in comparison occupations like licensed healthcare and teaching (−41%). Ads asking for juniors, trainees or new graduates fell 66% in the exposed group and 36% in the comparison group. Ads aren’t hires, the word junior can vanish from an ad while the job stays, and the business cycle, interest rates and the tech cycle moved over the same years. The report names those as strong alternative explanations. My conclusion is that AI may be one of several causes, which is reason enough to watch junior hiring closely. The Swedish version of this guide, Framtidens jobb och AI, has the full table.

Where do the big numbers come from?

Headlines about AI and jobs keep repeating the same four figures. Here is what each one measures.

Figure	Source	What it measures	What it doesn’t say
300 million jobs	Goldman Sachs, March 2023	Generative AI could expose the equivalent of 300 million full-time jobs worldwide to automation.	That 300 million people lose their jobs. With its own illustrative cut-offs, the report counts 7% of US jobs as likely to be substituted and 63% as complemented, and its baseline assumes most displaced workers find new jobs.
92 million jobs displaced	World Economic Forum, January 2025	Based on a survey of more than 1,000 employers, the WEF projects 170 million jobs created and 92 million displaced from 2025 to 2030, a net gain of 78 million.	That AI is the cause. The figures cover all the big trends, including technology, the economy, demographics and the green transition.
40% of jobs	IMF, January 2024	Almost 40% of global employment is exposed to AI, and about 60% of jobs in advanced economies may be impacted.	That all those jobs disappear. In advanced economies, the IMF expects roughly half of exposed jobs to benefit from AI. For the other half, AI may take over key tasks and lower demand, wages and hiring, and in extreme cases some jobs may disappear.
1 in 4 jobs	ILO, May 2025	25% of global employment is in occupations potentially exposed to generative AI, 34% in high-income countries.	Replacement. The ILO sees transformation as the most likely outcome.

I’d use these figures as orders of magnitude, never as forecasts of unemployment. Your own result from the five-step check tells you more about your job than any of them.

What you can do now

If you’re an employee

Use AI on your own tasks every week, with the tool your employer pays for. The four levels in my AI fluency framework show what the next step looks like in your role.
Take on what AI struggles with. Ask to lead a decision, a customer case or a difficult conversation you’d normally hand to someone else.
Ask your manager what’s planned. What does AI mean for my work this year? What’s expected of me? How much time do I get to learn?

If you’re a student or new to the job market

Check the ways in. Search job boards for the occupation together with junior, trainee, graduate or entry level. Few hits are a signal to plan another route, like an internship, an apprenticeship or a neighboring role.
Show you can use AI in the work. Do a typical task in the field with AI and write a few lines on what you changed in the draft and why. That shows judgment and says more than a line on your resume.
Get what AI can’t give you. Part-time jobs, internships and projects where you deal with customers, patients or colleagues build the skills that are hardest to replace.
Ask how new hires learn. In interviews, ask which tasks a new hire gets when AI does the simplest ones, and who coaches them.

If you’re a manager

Go through the team’s tasks. Do the exercise together and agree where the freed-up time goes. The 45-minute task sort is a good way to start.
Keep the entry-level jobs. The juniors you hire today are the seniors you need in ten years. When AI takes the simple tasks, new hires can learn by reviewing AI drafts with an experienced colleague.
Talk openly. Say what you know, what you don’t and what won’t change. My conversation guide has answers you can adapt, including to “Will AI take my job?”

If you work in HR

Plan the workforce from tasks. Run the exercise for your three largest job families, see where the hours move and plan transitions early for roles above 50%. My guide to strategic workforce planning goes further.
Track junior hiring. The share of new hires under 26, by job family and year, shows whether the ways in are closing without anyone deciding it.
Measure what AI changes. Track the hours saved on a few common tasks and what the time goes to instead.
Watch the rules if you operate in the EU. Under the EU AI Act, the high-risk rules for AI used in recruitment and decisions about workers apply from 2 December 2027. Employers must then inform workers’ representatives and the affected workers before using such a system at work.

For what AI means for HR’s own jobs, see Will AI take HR jobs? For data on how HR teams in Sweden use AI, see State of AI in HR.

Frequently asked questions

Which jobs will AI replace?

AI mostly takes over tasks in jobs built on text, code, numbers and conversations that follow a pattern, like data entry, customer service, translation, writing and programming. In US projections to 2035, the steepest declines include word processors and typists, telephone operators, data entry keyers and telemarketers.

Which jobs are safe from AI?

There are no AI-proof jobs, but jobs that combine physical work, licenses and legal accountability, trust and relationships, or judgment in unclear situations are the hardest to replace. Examples include roofers, home health aides, nurse practitioners and cooks. Low AI exposure doesn’t protect a job from other changes, so check demand too.

Will AI take my job?

Most likely it will take parts of it. How much depends on the share of your time that goes to tasks AI can draft or do, and on whether demand for your work grows when it gets cheaper.

Is it true that AI will replace 300 million jobs?

Not in the way it’s usually repeated. Goldman Sachs estimated in 2023 that generative AI could expose the equivalent of 300 million full-time jobs worldwide to automation. The same report expects most jobs to be complemented rather than substituted, and its baseline assumes most displaced workers find new jobs.

What jobs will be in demand in the future?

In the US, the BLS projects the fastest growth from 2025 to 2035 for nurse practitioners, solar photovoltaic installers, data scientists and wind turbine service technicians, with healthcare growing fastest overall.

Will AI cause mass unemployment?

There are no signs of it so far. The OECD found no signs of widespread displacement from AI at the industry level in its June 2026 outlook, and Stanford researchers see no economy-wide job losses in US payroll data. AI was the most-cited reason in US layoff announcements in the first nine months of 2026, but those 120,136 announced cuts are less than a tenth of a percent of US jobs. The clearest effect is that fewer young people are hired into the most AI-exposed jobs. Over the longer term, the uncertainty is large.

How do I know if my job is at risk from AI?

List your most common tasks and how much time they take, test which ones AI can make a good first draft of and work out the share of your time. Under 10% means small changes, 10 to 49% that AI becomes a tool in your job and 50% or more that the job is likely to change at its core.

Sources

Want to know which tasks AI will change in your organization?

I give keynotes and run workshops for leadership teams and whole organizations on which tasks AI is changing in your roles and what to do about it now. In my open sessions on AI in August and September 2026, 309 evaluation responses gave an average of 4.3 out of 5. If you’re reading for your own career, I write about AI, HR and the future of work in my newsletter, FullStack HR.

Talk about your organization