Will AI take my job? What changes at work, and what doesn't

Is AI going to replace me?

Probably not in one go, and probably not all of you. The researchers who study this mostly look at tasks, the individual things you do in a day, rather than whole jobs. A job is a bundle of tasks, and generative AI (tools that produce text, images or code on request) can do some of them far better than others.

That is why the International Labour Organization (ILO), the UN's agency for work, says that "transformation of jobs is the most likely impact" of generative AI. Most occupations, it notes, still include plenty of tasks that need a person.

How many jobs are actually exposed?

Quite a lot, depending on how you count. In its May 2025 update, the ILO estimates that about one in four workers worldwide are in an occupation with some exposure to generative AI. Only 3.3% of global employment falls into its highest-exposure group.

The International Monetary Fund (IMF) uses a broader measure of artificial intelligence in general. In January 2024 its managing director, Kristalina Georgieva, wrote that almost 40% of global employment is exposed, rising to about 60% in advanced economies. She added that roughly half of those exposed jobs in rich countries may gain from AI, while the other half could see lower demand, wages and hiring.

"Exposed" is the word to watch. It means a technology could do some of your tasks, not that your employer will use it, or that your job will go. The ILO says plainly that its figures "reflect 'exposure' to GenAI and not the actual impact on employment," and that cost, skills, infrastructure and social acceptance all slow adoption.

Which kinds of work are most and least exposed?

According to the ILO, clerical work is the most exposed: data entry clerks, typists, bookkeeping clerks and general office clerks. Some digital professional jobs, such as financial analysts and web and multimedia developers, have also moved up its list.

The least exposed work involves the physical world. The ILO's experts gave lower scores to trades, machine operators and farm work because of "the practical limitations of automating tasks that require physical manipulation or manual dexterity."

One pattern surprises people. Exposure is higher in better-paid office jobs than in manual ones. Statistics Canada found that jobs more exposed to AI are more likely to be well paid, full time and permanent. The ILO also finds that women are more likely than men to work in the most exposed jobs: 4.7% of employed women versus 2.4% of men worldwide, and the gap is wider in high-income countries.

What has actually happened to jobs so far?

So far, the big picture is calmer than the headlines.

• No wave of unemployment yet. A September 2025 study from the University of Chicago, using US government survey data, found the effect of large language models on unemployment in exposed occupations was centred around zero, with earnings slightly up.
• Canada looks similar. Statistics Canada reported in January 2026 that employment generally grew between November 2022 and December 2025, whatever a job's exposure to AI. It cautions that some trends predate AI and could reflect other things, such as post-pandemic adjustments and immigration.
• The Yale Budget Lab told Fortune in February 2026 that its data shows stability rather than economy-wide disruption. "It just doesn't seem like there's major macroeconomic effects here," said its researcher Martha Gimbel.

The clearest warning sign concerns young people. Economists at the Stanford Digital Economy Lab, using payroll data from the firm ADP, found that by June 2026 employment for 22-to-25-year-olds in the most AI-exposed occupations was 19% below where it would have been had it kept pace with less exposed jobs. Software development and customer service are among their examples. Older workers in the same jobs showed no comparable gap.

The authors call these early "canaries in the coal mine," not proof that AI caused the decline. Their data overrepresents large firms and shows a bigger effect than national surveys do. Statistics Canada saw a related pattern on a smaller scale: coding jobs grew overall, but coding professionals under 30 stagnated.

Are companies really firing people because of AI?

Some say they are. The US outplacement firm Challenger, Gray & Christmas counts the reasons employers give when they announce job cuts. Its August 2026 report says AI was cited for 116,175 announced US cuts so far this year, about 22% of the total, making it the leading reason year to date.

Treat stated reasons with some care. An Oxford Economics report, cited by Fortune, suggested that "some firms are trying to dress up layoffs as a good news story rather than bad news." Critics call this "AI washing." The reverse also happens: when Amazon announced 16,000 corporate cuts in January 2026, Fortune reports, the company put them down to reducing bureaucracy, not AI.

Does AI actually make people better at their jobs?

Sometimes, and not evenly. A large study of 5,179 customer support agents, published in the Quarterly Journal of Economics in 2025, found that an AI assistant raised the number of issues resolved per hour by 14% on average. Novice and lower-skilled agents improved by 34%; the most experienced barely changed.

Expert work is a different story. In an early-2025 experiment by the research group METR, experienced software developers took 19% longer on real tasks when allowed to use AI tools, even though they believed the tools had sped them up. METR's February 2026 follow-up suggests newer tools probably help more, but calls its own data "very weak evidence" of how much.

The lesson: AI tends to help most where it fills a gap in experience, and its benefits are easy to overestimate. It also makes mistakes, called hallucinations, so somebody still has to check the work.

What do employers and workers expect?

The World Economic Forum's Future of Jobs Report 2025, based on a survey of over 1,000 companies, expects 170 million jobs to be created and 92 million displaced by 2030, a net gain of 78 million. That is a forecast from employers, not a measurement. Employers expect nearly 40% of the skills needed on the job to change. 41% plan to reduce staff where AI can do the work, and 77% plan to retrain their people.

Workers are uneasy. In a Pew Research Center survey of US workers in October 2024, 52% said they were worried about how AI will be used at work in future, and 36% were hopeful. Actual use is rising but still modest: by September 2025, 21% of US workers told Pew that at least some of their work is done with AI.

What can I do about it?

• Learn the tools your field actually uses. Not every chatbot on the market, just the two or three your colleagues, clients or competitors use. Try them on tasks you know well, so you can see where they help and where they go wrong. Our guide to writing a good prompt is a quick start.
• Lean into the parts of your job that need you. Judgement, relationships, responsibility and physical skill are, on current evidence, the hardest tasks to hand over.
• Know your employer's AI policy. Ask whether there is one, which tools are allowed, and what you may paste into them. Company data in a public chatbot can get you into trouble. See your data and AI chatbots.
• Ask about training. Under the EU's AI Act, organisations that use AI systems must take measures to support AI literacy among their staff. Since the July 2026 "Digital Omnibus" amendments, that duty is softer than first drafted, according to the law firm Lewis Silkin, but it still applies.
• Know your rights if you are in the EU. The AI Act lists AI used to recruit people, filter applications, assess candidates, decide on promotions or dismissals, allocate tasks, or monitor performance as "high-risk". Employers using such systems must inform workers' representatives and the affected workers, and people subject to decisions made with them must be told. Using AI to infer the emotions of workers has been banned since 2 February 2025, except for medical or safety reasons.
• Note the timing. After the Digital Omnibus, the main high-risk rules for these employment systems apply from 2 December 2027, not August 2026 as originally planned. Your national data protection and labour laws already apply today. Our EU AI Act explainer covers the rest.
• If you are early in your career, be aware that the evidence points to entry-level openings in exposed office and software roles as the place where change shows up first. Experience that only a person can gain on the job, with clients, on site or in a team, is worth seeking out.