AI Pulse Check
← Back to blog

What Jobs Will AI Replace? What the Research Says (2026)

By Ethan Halfhide · Updated November 21, 2026

The jobs AI is most likely to replace are built around routine information tasks: data entry, cashier and ticket work, administrative assistance, bank tellers, telemarketing, basic customer service and some claims processing. Most research measures exposure, not replacement, and so far the evidence points to tasks being automated inside jobs far more often than whole jobs disappearing.

This guide pulls together what the major studies actually say: the World Economic Forum, the International Labour Organization (ILO), the US Bureau of Labor Statistics (BLS), Goldman Sachs, OpenAI and the University of Pennsylvania, Microsoft, Anthropic, Stanford and the Yale Budget Lab. Then it turns that into decisions for a small business owner: which tasks to automate, which roles to protect, and how to hire.

The short answer

  • Most exposed: clerical and office support work, customer service, data entry, cashiers, bank tellers, telemarketers, proofreaders, credit analysts, claims adjusters and auto damage appraisers, medical transcriptionists, and parts of programming, writing and sales.
  • Least exposed: work that needs a body in a place, or trust and judgment in person: nursing assistants and home health aides, construction, grounds maintenance, farm work, delivery, and many healthcare and care roles.
  • What is happening now: the broad US labor market has not shown a clear AI shock, but several 2025 and 2026 studies find weaker hiring of workers aged 22 to 25 in the most exposed occupations.
  • For a small business: treat AI as a way to take hours out of specific tasks. Very few small teams have a role that is 100% routine information work.

Exposure is not replacement: what each study measures

Headlines often turn "exposed to AI" into "replaced by AI." The studies are more careful.

Study (year)What it measuresHeadline figure
WEF Future of Jobs Report 2025A survey of over 1,000 employers about expected job changes from 2025 to 2030170 million jobs created, 92 million displaced, a net increase of 78 million
ILO Working Paper 140 (May 2025)Which occupations have tasks generative AI could perform, worldwide"one in four workers are in an occupation with some GenAI exposure"; 3.3% of global employment is in the highest exposure category
"GPTs are GPTs", Eloundou, Manning, Mishkin and Rock (2023)Share of tasks in US jobs that large language models could affect"around 80% of the U.S. workforce could have at least 10% of their work tasks affected"; "approximately 19% of workers may see at least 50% of their tasks impacted"
Goldman Sachs (April 2023)Exposure of jobs to automation by generative AI"roughly two-thirds of U.S. occupations are exposed to some degree of automation by AI"; the equivalent of 300 million full-time jobs exposed globally
Goldman Sachs Research (August 2025)Estimated displacement if AI is widely adoptedCould "displace 6-7% of the US workforce if AI is widely adopted", with a range of 3% to 14% under different assumptions
BLS Monthly Labor Review (February 2025)How BLS built AI into its 2023 to 2033 occupation projectionsCustomer service representatives projected at minus 5.0%, while software developers are projected at plus 17.9%
BLS Employment Projections 2025 to 2035 (August 2026)Official 10-year US job projectionsOffice and administrative support is projected to decline 4.0%, with AI-powered automation named as a cause
Microsoft, "Working with AI" (2025)How well AI performs work activities, based on "a dataset of 200k anonymized conversations with Microsoft Bing Copilot"Media and communication workers score highest (0.38); forest and conservation workers lowest (0.03)
Anthropic Economic Index (January and March 2026)What people actually use Claude for, mapped to occupational tasks"About 49% of jobs have seen at least a quarter of their tasks performed using Claude" (March 2026)

Three points matter when you read any of these:

  1. Exposure means AI could do or speed up a task. It does not say whether employers will cut staff, hire fewer people, or keep the same team and do more.
  2. Usage data is not a forecast. Anthropic and Microsoft measure what people already do with their AI tools. That shows where AI is being applied, not where jobs will disappear. Microsoft's authors write that "AI's effects on employment and wages will depend on hard-to-predict business decisions."
  3. Most researchers expect transformation first. The ILO says "transformation of jobs is the most likely impact of GenAI," and Goldman Sachs wrote in 2023 that most jobs are "more likely to be complemented rather than substituted by AI."

Roles most exposed to AI

Each role below is flagged as most exposed, declining or at risk in at least one major source.

RoleWhat the research saysSource
Data entry clerks and keyersAmong the fastest-declining roles; Anthropic's observed exposure measure, which combines AI capability with real Claude usage, rates data entry keyers 67% coveredWEF Future of Jobs Report 2025; Anthropic, "Labor market impacts of AI" (March 2026)
Cashiers and ticket clerksAmong the largest declines in absolute numbersWEF Future of Jobs Report 2025
Administrative assistants and executive secretariesAmong the largest declines in absolute numbersWEF Future of Jobs Report 2025
Office and administrative support (group)Projected to fall 4.0% (752,100 jobs) from 2025 to 2035, partly due to "automation tools, including those powered by AI"BLS Employment Projections (August 2026)
Bank tellers and postal service clerksAmong the fastest-declining rolesWEF Future of Jobs Report 2025
Customer service representativesProjected minus 5.0% from 2023 to 2033; named among occupations most at risk; second most exposed on Anthropic's observed exposure measureBLS Monthly Labor Review (2025); Goldman Sachs (2025); Anthropic (March 2026)
Telemarketers, proofreaders and copy editorsNamed among occupations most at risk of displacementGoldman Sachs (2025)
Credit analystsProjected minus 3.9% from 2023 to 2033; also named as at riskBLS Monthly Labor Review (2025); Goldman Sachs (2025)
Claims adjusters and auto damage appraisersProjected minus 4.4% and minus 9.2% from 2023 to 2033BLS Monthly Labor Review (2025)
Medical transcriptionistsProjected minus 4.7% from 2023 to 2033BLS Monthly Labor Review (2025)
Computer programmersTop of Anthropic's observed exposure measure at 75% task coverage; named at risk by Goldman SachsAnthropic (March 2026); Goldman Sachs (2025)
Accountants and auditors, legal and administrative assistantsNamed among occupations most at riskGoldman Sachs (2025)
Media and communication workersHighest AI applicability score (0.38) of any occupation groupMicrosoft, "Working with AI" (2025)
Sales representatives (services)Second highest applicability score (0.35); BLS projects sales occupations to fall 1.4% from 2025 to 2035Microsoft (2025); BLS (August 2026)
Graphic designersAdded to the WEF list of fastest-declining rolesWEF Future of Jobs Report 2025 press release

Two cautions. First, an exposed job is not automatically a shrinking job: BLS projects software developers to grow 17.9% and paralegals 1.2% from 2023 to 2033 even though AI exposure is high in both. Second, clerical work stands out in almost every source. The ILO found in 2023 that "nearly a quarter of tasks" in clerical work were highly exposed and "more than half of tasks" had medium exposure, and its 2025 update says "clerical occupations continue to have the highest exposure levels."

Jobs least likely to be replaced

These roles involve physical work in unpredictable settings, personal care, or high accountability.

RoleWhat the research saysSource
Home health aides and nursing assistantsAmong the lowest AI applicability scores (0.04)Microsoft (2025)
Grounds maintenance workersApplicability score of 0.04Microsoft (2025)
Construction workers"Other construction and related workers" score 0.06; construction workers are among the largest expected job gainsMicrosoft (2025); WEF Future of Jobs Report 2025
Farmworkers, delivery drivers, food processing workersExpected to see the largest growth in absolute termsWEF Future of Jobs Report 2025
Nursing professionals and secondary school teachersExpected to see significant increasesWEF Future of Jobs Report 2025 press release
Healthcare and social assistance (sector)Projected to grow 9.5%, adding over 2.2 million jobs, driven by the aging population and more chronic health conditionsBLS Employment Projections (August 2026)
Personal financial advisorsProjected plus 17.1% from 2023 to 2033 despite robo-advisorsBLS Monthly Labor Review (2025)
Air traffic controllers, chief executives, pharmacists, clergyNamed among occupations least at riskGoldman Sachs (2025)
Data scientistsProjected plus 34.6% from 2025 to 2035BLS Employment Projections (August 2026)

The studies do not always agree. Goldman Sachs lists radiologists among the least at risk, while Anthropic's January 2026 report says radiologists "move up" on its effective AI coverage measure because the tasks Claude handles well, interpreting diagnostic images and preparing reports, are their most time-intensive work. A job can contain many tasks AI handles well and still depend on licensed judgment that keeps a person in the role.

What has actually happened so far

  • No broad shock yet. The Yale Budget Lab (October 2025) concluded that "the broader labor market has not experienced a discernible disruption since ChatGPT's release 33 months ago."
  • No rise in unemployment for exposed workers. Anthropic's March 2026 labor market study states: "We find no systematic increase in unemployment for highly exposed workers since late 2022."
  • Weaker entry-level hiring. The same Anthropic study finds "tentative evidence that hiring into those professions has slowed slightly for workers aged 22-25." Stanford's Digital Economy Lab (Brynjolfsson, Chandar and Chen, revised August 12, 2026), using ADP payroll data through June 2026, finds employment of young workers in AI-exposed occupations is "19% below" where it would be had it kept pace with less exposed peers, with no comparable gap for experienced workers. It also finds declines are concentrated where AI "primarily substitutes for human tasks."
  • Goldman Sachs expects a transition, not a collapse. Its August 2025 note estimates unemployment "will increase by half a percentage point during the AI transition period" and adds "we remain skeptical that AI will lead to large employment reductions over the next decade."
  • Employers are planning for change. In the WEF survey, 41% of employers say they "plan to reduce their workforce as AI automates certain tasks" (the report digest gives 40%), while 77% plan to upskill their workforce and two-thirds plan to hire talent with specific AI skills.

So far, the first effect shows up in junior hiring for routine information roles, not in mass layoffs.

What it means for a small business

Small businesses use AI less than large ones. The US Census Bureau's Business Trends and Outlook Survey found overall business AI use "hovered between 17% and 20%" from December 2025 to May 2026, with "less than 20% of firms with four or fewer employees" using it, compared with 37% of firms with at least 250 employees. Yet the most exposed tasks are the ones that eat a small team's week.

Automate tasks, not roles

In a small business, one person often handles phones, inbox, invoicing and scheduling. AI is good at parts of that work, not all of it. Start with the task types most exposed across studies:

  • Answering repeat customer questions and routing the rest
  • Data entry from forms, receipts and invoices into your systems
  • Scheduling, reminders and follow-up messages
  • First drafts of emails, quotes, product descriptions and social posts
  • Summarizing calls, documents and long email threads
  • Basic bookkeeping categorization (with a person reviewing)

Our guide to the best AI tools for small business covers tools for these jobs. For multi-step work that runs across your apps, such as triaging an inbox and updating a CRM, read about AI agents for business.

Roles to keep and invest in

Keep people on work where the research shows low exposure or where mistakes are expensive: in-person service and skilled trades, sales conversations that close deals, customer escalations and complaints, anything regulated or licensed, and final review of anything AI drafts. Move any time AI frees up to follow-up calls, account management or quality control.

Reskilling

The WEF report says 39% of workers' existing skill sets "will be transformed or become outdated over the 2025-2030 period." For a small team, reskilling can be simple: teach each person to use AI on their own repetitive tasks, to check its output, and to know what customer data should never go into a public tool.

Hiring

Before you post a junior admin or support role, list the tasks it covers and mark which ones AI could take on now. You may still need the hire, but the job description may shift toward judgment, customer contact and supervising tools. That matches what the Stanford and Anthropic data show employers doing: slower hiring into entry-level, routine information roles rather than cuts to experienced staff.

A five-step task audit

  1. List the recurring tasks in your business for one week, with rough hours each.
  2. Mark each task as routine information work, judgment, physical or relationship.
  3. Pick the two routine information tasks that take the most hours.
  4. Run a small pilot with one tool for 30 days, with a person checking every output.
  5. Keep, adjust or drop it based on time saved and error rate, then pick the next task.

If you want a fuller plan with goals, budget and rules for staff, follow our AI adoption strategy framework.

Will AI take my job?

If you are asking for yourself rather than your business, look at your own week, not your job title. The "GPTs are GPTs" authors estimate about 19% of US workers could see half or more of their tasks affected, which means most people will see a share of their tasks change, not all of them. Anthropic's June 2026 survey of about 9,700 Claude users (which Anthropic notes is not representative of the general population) found "10% rated losing their own jobs as likely or very likely," and that early-career workers expressed the most concern about job loss. The most durable move is to become the person who uses AI well on the routine parts of your job and owns the parts that need judgment, relationships or physical skill.

What we could not verify

  • Exact timelines. None of these studies says when a given job will be replaced. The "GPTs are GPTs" authors state: "We do not make predictions about the development or adoption timeline of such LLMs."
  • Job-title-level lists from Microsoft. The Microsoft paper publishes occupation group scores in its main table; we used those group scores and did not quote individual job rankings shown only in its charts.
  • Small business specific displacement data. We found no official dataset that measures job losses from AI in firms under 20 employees.

FAQ

What jobs will AI replace first?

The roles the research flags earliest are routine information jobs: data entry clerks, cashiers and ticket clerks, administrative assistants, bank tellers, postal clerks, telemarketers and basic customer service. The WEF Future of Jobs Report 2025 names postal service clerks, bank tellers and data entry clerks among the fastest-declining roles, and BLS projects office and administrative support to fall 4.0% from 2025 to 2035.

What jobs can AI not replace?

No study says any job is fully immune, but the lowest exposure is in physical, in-person and care work. Microsoft's 2025 study gives home health aides, nursing assistants and grounds maintenance workers some of the lowest AI applicability scores, and the WEF expects the largest absolute growth in farmworkers, delivery drivers, construction workers, salespersons and food processing workers.

Is being exposed to AI the same as losing your job?

No. Exposure means AI could perform or speed up some of your tasks. The ILO says transformation of jobs is the most likely impact, and BLS projects some highly exposed jobs, such as software developers, to keep growing because demand rises along with productivity.

What jobs will be gone by 2030 because of AI?

No major report predicts that whole occupations will vanish by 2030. Employers in the WEF survey expect 92 million jobs to be displaced and 170 million to be created between 2025 and 2030, with the steepest declines in clerical and secretarial roles.

Can AI replace teachers?

The research does not point that way. The WEF expects secondary school teachers to see significant job increases by 2030. Anthropic's January 2026 report found educational instruction is the second-largest category of Claude.ai use, at 15%, but that measures how people use AI, not job losses.

Should a small business replace staff with AI?

Usually the better move is to replace tasks, not people. Use AI on routine information work, keep a person reviewing its output, and move the time saved to sales, service and quality. Revisit hiring plans for junior routine roles before you post them.

Is AI already causing job losses?

Not at the level of the whole US economy. The Yale Budget Lab found no discernible disruption in the broad labor market, and Anthropic found no systematic rise in unemployment for highly exposed workers. The clearest signal so far is weaker employment and slower hiring of workers aged 22 to 25 in AI-exposed occupations.

How this guide was put together

This is not our own study. We compared published research from the World Economic Forum, ILO, BLS, Goldman Sachs, OpenAI and University of Pennsylvania researchers, Microsoft, Anthropic, Stanford's Digital Economy Lab, the Yale Budget Lab and the US Census Bureau. Every figure is quoted from the original report or the publisher's own summary, with the report name and year, and checked against the source in November 2026. Where studies measure different things (survey expectations, task exposure, real usage, payroll data), we say which is which.

Sources