AI is replacing tasks before it replaces whole jobs, and the US roles losing the most ground are customer service representatives, data entry clerks, bank tellers, bookkeepers, entry-level programmers, paralegals, and routine writing and translation work. If you want a direct answer to which jobs AI is replacing, the evidence points to that list.
The longer answer matters more than the list. The same tools that shrink one team often grow another, and so far the effect shows up mostly as slower hiring, with fewer mass layoffs.
This guide draws on named research from the World Economic Forum, Goldman Sachs, Stanford, and Microsoft, plus public company statements. It ends with a short test you can run on your own job.
What the Research Says About AI and Jobs
Goldman Sachs estimated in 2023 that generative AI could expose the equivalent of 300 million full-time jobs worldwide to automation. Its analysts also found that about two-thirds of US occupations have some exposure, though for most of them AI would handle only a portion of the work.
The World Economic Forum’s Future of Jobs Report 2025 surveyed employers and projected that 92 million jobs will be displaced by 2030 while 170 million new ones are created, a net gain of 78 million. The fastest-shrinking roles on its list include postal service clerks, bank tellers, data entry clerks, cashiers, and administrative assistants. Those figures cover all workplace change, not AI alone.
Real-world layoffs give a second signal. The outplacement firm Challenger, Gray & Christmas has tracked tens of thousands of announced US job cuts since 2023 where employers named AI as a reason, a small share of total cuts but a growing one.
The 8 Jobs Losing Ground to AI First
These eight show up repeatedly across research on what jobs AI is replacing and in company announcements. The order is not a ranking of exact job losses, since no agency publishes a clean count of AI-caused layoffs.
1. Customer Service Representatives
This is the clearest case. Salesforce CEO Marc Benioff said in 2025 that AI agents let the company cut its support team from about 9,000 to about 5,000 people. Klarna made a similar claim in 2024, saying its AI assistant handled the work of roughly 700 agents, then began bringing human staff back in 2025 after quality concerns.
The US Bureau of Labor Statistics projects customer service representative employment to decline this decade. Complex, emotional, or high-value cases still land with humans.
2. Data Entry Clerks and Administrative Assistants
Both appear near the top of the World Economic Forum’s list of fastest-declining roles. The work is repetitive, rule-based, and already digital, which makes it easy for software to absorb. Scheduling, form processing, invoice capture, and basic email triage are now handled by AI tools.
Assistants who manage people, judgment calls, and exceptions are safer than those who mostly process paperwork.
3. Bank Tellers and Cashiers
The WEF lists both among its fastest-declining jobs through 2030. Mobile deposits, self-checkout, and app-based ordering did much of the work, and AI-driven fraud checks and voice assistants continue the trend. Those changes are about automation more broadly than generative AI, but employers now bundle them together in workforce plans.
Roles that survive tend to shift toward advising and problem-solving at the branch.
4. Bookkeepers and Junior Accountants
Software now categorizes transactions, reconciles accounts, and drafts routine reports with little human input. For a small business, a single owner can now run tools that once needed a part-time bookkeeper. The work that remains leans toward review, tax judgment, and client conversations.
5. Entry-Level Software Developers
This is the most debated case. A 2025 Stanford Digital Economy Lab paper by Erik Brynjolfsson and coauthors, using payroll data from ADP, found a 13% relative employment decline for workers aged 22 to 25 in the most AI-exposed occupations, with software development and customer service among the clearest examples. Older developers in the same fields saw stable or rising employment. AI coding tools handle boilerplate and bug fixes that juniors once learned on.
6. Writers, Translators, and Interpreters
Microsoft researchers analyzed 200,000 anonymized Copilot conversations in 2025 and ranked occupations by how well AI can do their tasks. Interpreters and translators topped the list, followed by historians, passenger attendants, sales representatives, and writers and authors. Freelance and agency work for routine copy, product descriptions, and basic translation is where the pressure is highest.
Editors, strategists, and specialists with original reporting hold up better.
7. Paralegals and Legal Research Staff
Document review, contract summaries, and case-law searches are tasks language models handle quickly. Law firms increasingly use AI tools to speed up first-pass review, which can cut the hours billed to junior staff. Courts have also sanctioned lawyers who filed AI-generated briefs with fake citations, so a human check remains mandatory.
That keeps demand for trained reviewers alive, although fewer of them per case.
8. Telemarketers and Outbound Sales Reps
Voice AI can now place calls, follow a script, and book appointments at a scale no call center floor can match. The FCC ruled in February 2024 that AI-generated voices in robocalls fall under the Telephone Consumer Protection Act, which limits how far this can go legally. Outbound roles built on cold-calling volume are the most exposed. Relationship-based sales, where trust and negotiation drive the deal, are much harder to replace.
Jobs AI Is Changing Faster Than Replacing
Many jobs sit in the middle. Goldman Sachs found that most exposed occupations would see only a slice of their tasks automated, which points to rewritten job descriptions more than empty desks.
Marketing managers, financial analysts, radiologists, and teachers all fit this pattern. An analyst who once spent Monday pulling numbers into a spreadsheet now spends it questioning what an AI summary got wrong. The job still exists, but the mix of tasks has moved toward judgment and away from assembly.
Amazon CEO Andy Jassy told employees in a June 2025 memo that generative AI would shrink the company’s corporate workforce over the next few years, even as other roles grow. Statements like that matter because they signal how large employers plan headcount, not just what tools can do in a lab.
The WEF report shows the other side of the ledger. Big data specialists, fintech engineers, AI and machine learning specialists, and software developers lead its list of fastest-growing roles.
Why Entry-Level Workers Feel It First
Entry-level jobs are built from the exact tasks AI does well: summarizing documents, drafting first versions, cleaning data, and answering standard questions. That work used to be how new hires learned the trade.
Anthropic CEO Dario Amodei warned in a May 2025 interview with Axios that AI could eliminate half of entry-level white-collar jobs within one to five years and push unemployment to 10% to 20%. That is a forecast from an AI company leader, not a measured result. The Stanford study is the closest measured evidence so far, and it covers only a few years of data.
Two things can be true at once. Overall US unemployment can stay low while a narrow group of young workers in exposed fields struggles to land a first job. That gap is easy to miss in national averages and easy to overstate in headlines.
For hiring managers, the trade-off is real. Teams that skip junior hires today may find they have no bench of mid-level people in three years. Some companies may respond by redesigning junior roles around review and supervision of AI output.
Jobs That Are Hardest for AI to Replace
Work that happens in physical, unpredictable spaces is the hardest to automate. Electricians, plumbers, roofers, nurses, and physical therapists move through different homes, bodies, and problems every day. Microsoft’s occupation ranking puts roles like dishwashers, roofers, and floor sanders near the bottom of AI applicability. That ranking measures chatbot usefulness, so it says little about future robotics.
Regulated professions add another layer of protection. Licensing rules, malpractice liability, and safety standards slow the handoff of medical, legal, and engineering decisions to software. Health and legal decisions in particular stay with licensed humans, even when AI drafts the first pass.
Where Job Growth Is Still Coming From
The WEF report expects some of the largest absolute gains through 2030 in farm work, delivery driving, software development, construction, and nursing.
Software development appears on both the growth and at-risk lists because the field is splitting. Demand for people who can build and supervise AI systems is rising, while demand for routine coding is under pressure. The same split shows up in writing, design, and analytics.
Wages tell a similar story. PwC’s 2025 Global AI Jobs Barometer reported a wage premium for jobs that require AI skills.
A Three-Question Test for Your Own Job
Job titles hide the real picture, so break your week into tasks and ask three questions about each one.
- Does the task happen entirely on a screen, with clear inputs and outputs? Anything that fits that description is a candidate for automation.
- Would a wrong answer be cheap to catch and fix? AI makes confident mistakes, so tasks with low error costs and easy review get automated first.
- Does anyone pay for your name, license, or relationship rather than the output alone? If yes, the task is safer.
If most of your hours score high on the first two and no on the third, your role deserves attention. If you use AI to finish those tasks faster, you are usually better placed than a colleague who ignores it.
What Workers and Employers Can Do Now
1. For Workers
Start by learning the tools your field already uses. A paralegal who can check AI research for fake citations, or a bookkeeper who audits automated categorization, offers something a software subscription cannot. Employers tend to keep people who make the output more reliable.
Next, move toward the parts of your job that involve people, exceptions, and accountability. The BLS Occupational Outlook Handbook lists projections by occupation, which helps when you compare career paths.
2. For Employers
Treat AI as a task tool first. Map which tasks in each role can be automated, then decide whether that frees people for higher-value work or removes the role. The second choice looks efficient on paper, but Klarna’s decision to bring back human agents shows it can backfire.
How to Read the AI Job Loss Headlines
Exposure is not the same as replacement. When Goldman Sachs or Microsoft says a job is exposed, it means AI can do part of the tasks, not that employers have cut the job. Adoption depends on cost, accuracy, legal risk, and how long a company takes to rebuild its workflows.
Layoff announcements can also overstate the cause. A company under pressure to cut costs may credit AI because it sounds forward-looking, even when interest rates, overhiring, or weak sales did most of the work. Challenger’s tracking is based on the reasons employers give, which is useful but not a controlled measurement.
Forecasts deserve the most caution. The WEF numbers come from employer surveys about plans, and executive predictions come from people with a stake in the technology. Measured studies, like the Stanford payroll analysis, are narrower but more reliable, and they need to be repeated as new data arrives.
A useful habit is to revisit the BLS employment data and the Challenger job cut reports each quarter, since the answer to what jobs AI is replacing keeps changing.
