In short: AI is more likely to replace tasks than whole jobs in the near term, but roles made up mostly of routine information work face real pressure. These include some customer service, administrative and junior office positions. Evidence so far shows early signs among young workers, not mass unemployment. Forecasts vary widely, so assess your own tasks, not job titles.
Lists of doomed professions make good headlines. They’re usually less helpful than they seem, because they rely on how much of a job AI could theoretically do. A job can be highly exposed and still grow. “Could do the task” and “will eliminate the job” are separate claims.
Roles under the most pressure
Researchers tend to flag jobs built around text, information and repeatable rules.
| Role | Why it’s exposed | Why it may not vanish |
|---|---|---|
| Customer service representatives | Chatbots handle common questions | Complex cases, angry customers, liability |
| Administrative and office support | Scheduling, data entry, drafting | Exceptions and coordination |
| Data entry and claims processing | Rule-based document handling | Errors can be expensive; oversight needed |
| Basic translation and transcription | Quality is good on routine material | Legal, medical and literary work needs review |
| Routine content writing | Fast cheap drafts | Original reporting and expertise |
| Junior coding and analysis | Repetitive code can be automated | Someone has to check and own the result |
| Bookkeeping | Categorization and reconciliation | Judgment, advice and accountability |
The US Bureau of Labor Statistics, in its 2024 to 2034 projections, expects automated systems including AI to reduce employment of office and administrative support workers, and says AI in sales may limit demand for sales workers. It also notes that uncertainty about AI’s effects is very high, and it doesn’t assume a broad jump in productivity.
Stanford’s Digital Economy Lab offers the best-known early evidence. Erik Brynjolfsson, Bharat Chandar and Ruyu Chen reported a 13% relative decline in employment for workers aged 22 to 25 in the most AI-exposed occupations since generative AI spread, even after controlling for firm-level shocks. Older workers in the same occupations saw increases of 6% to 9%. In the study’s listing, customer service representatives, software developers, accountants and receptionists were among the most exposed to automation. It’s a working paper, and young workers may be affected by other factors as well.
Why the picture isn’t settled
Other respected analyses find much less so far.
- Yale’s Budget Lab tracked changes in the occupational mix and unemployment duration after ChatGPT’s release, and described changes as too slow to signal a major shift. Its director told Fortune that no major macroeconomic effects were visible in the data.
- The Chicago Fed found that employment in the occupations most exposed to AI rose 4.05% from 2019 to 2024, while less exposed occupations trimmed payrolls. Wages rose across the board. Note that this period mostly predates the widest use of the tools.
- Goldman Sachs went the other way in forecasting, raising its estimate in June 2026 to above 9% of US jobs at risk, about 15 million workers, over around a decade. It estimated a peak effect on unemployment of under 1% if the losses are spread out and most displaced workers find new work within a year.
Layoff announcements also need care. Oxford Economics, citing Challenger, Gray & Christmas, found that about 55,000 US job cuts in the first 11 months of 2025 were attributed to AI, 4.5% of reported cuts. The share has since jumped: Challenger counted 121,401 AI-attributed cuts from January to October 2026, roughly a fifth of all announced cuts. Some analysts suspect “AI washing,” in which companies blame AI for cuts driven by weak demand or past over-hiring because it sounds better to investors.
Then there’s reversal. Klarna froze hiring and said AI was doing the work of 700 agents, then began hiring human agents again in 2025 after its CEO acknowledged a loss of quality.
How replacement really tends to happen
Whole occupations seldom disappear overnight. Pressure usually arrives more quietly:
- Fewer entry-level openings. Companies stop backfilling junior roles, as a smaller team with AI can cover the routine work.
- Larger workloads. The same staff handle more, so headcount stops growing.
- Tasks move. A role shifts toward checking, exceptions and client contact.
- Pay splits. Those who use the tools well may command more, while routine work gets cheaper. On Upwork, per-contract earnings in generative AI and creative production fell 13% in the past year, even as contract starts rose 90%, while more complex AI-augmented work saw earnings rise 45%.
The first point deserves particular attention. If companies hire fewer juniors, it becomes harder to gain the experience needed for senior work. Employers may eventually feel that too, but it’s a real problem right now for young job seekers.
Assess your own job
Job titles hide too much. Break your job down into tasks, and estimate your weekly hours for each.
| Question | Higher risk | Lower risk |
|---|---|---|
| Is the work mainly digital? | Yes, all on a screen | Physical presence required |
| Are the rules clear? | Standard procedures | Messy, novel situations |
| Is it easy to check? | Output can be quickly verified | Results take time and judgment to assess |
| Does trust matter? | Buyers don’t care who did it | Relationships drive the decision |
| Who carries responsibility? | Errors are cheap | Someone must sign off and be accountable |
| Is demand elastic? | Cheaper output doesn’t mean more buying | Lower costs could bring more demand |
That last row is often missed. When a service becomes cheaper, people may buy far more of it, which can protect or expand employment. This has happened with past technologies, though nobody can guarantee it will repeat here.
What you can do
- Use the tools on your own work, starting with our guide to the best AI tools for productivity. People who understand the limits are harder to replace with them.
- Shift toward what’s harder to automate: client relationships, judgment, physical presence and responsibility for results. Our article on jobs safe from AI looks at what those roles have in common.
- Build a cushion and options. Savings and an updated network give you time to respond. See our guide to career development, and how to change careers if your assessment worries you.
- Be skeptical of certainty. Neither “nothing will change” nor “every office job is gone by 2030” has solid evidence behind it.
The bigger context is in our guide to AI and the future of work.
Frequently asked questions
Will AI replace software developers?
Current evidence suggests changes in how the work is done, along with pressure on entry-level hiring, rather than the end of the profession. The Stanford study flagged young developers as exposed. Yet BLS still expects strong demand tied to AI and data. Experienced developers who can specify, review and own systems look better placed than those who write only routine code.
When will AI take most jobs?
No one knows, and there is no credible consensus that it will. Forecasts differ widely, from no visible macro effect today to millions displaced over a decade. Earlier predictions of mass automation often overestimated the speed. Plan for uncertainty by building adaptable skills and savings, instead of betting on any specific date.
Are graduates hit hardest?
So far, some evidence points that way. The Stanford research found the decline concentrated among workers aged 22 to 25 in exposed fields. Others argue that other factors, such as slower hiring after the pandemic and higher interest rates, explain much of the weakness. Graduates can respond by gaining practical experience early and learning to use these tools.
