How AI Will Affect Jobs: Evidence, Pros, Cons and What Comes Next

How will AI affect jobs? The short answer

AI exposure means that an occupation contains tasks AI could affect. An exposed job is changed, not necessarily replaced ^[5]. That distinction explains why large headline numbers have not turned into mass unemployment.

IMF research, restated by Managing Director Kristalina Georgieva in October 2026, finds that roughly 40% of jobs worldwide are exposed to AI, rising to about 60% in advanced economies ^[1]. A 2025 index from the International Labour Organization (ILO) and the Polish research institute NASK finds that one in four jobs worldwide is potentially exposed to generative AI ^[5]. Generative AI is AI that produces new text, images, code or other content from a prompt.

Most of that exposure points to change inside jobs rather than elimination. The ILO–NASK authors conclude that few jobs consist of tasks that current AI can fully automate, so job transformation is the most likely impact ^[5]. The World Economic Forum (WEF) projects a net gain of 78 million jobs worldwide between 2025 and 2030 ^[2]. That figure, however, combines technology, demographics, geoeconomic tensions and economic pressures, not AI alone ^[2].

What has happened so far? US payroll and survey data through mid-2026 show no widespread, economy-wide job displacement linked to AI ^[3][4]. The clearest warning sign is narrower. Employment of 22–25-year-olds in highly AI-exposed occupations has fallen well behind that of their peers in less-exposed jobs, mainly because companies are hiring fewer of them ^[3].

How many jobs will AI replace, and how many will it create?

The most-quoted forecast is the WEF's Future of Jobs Report 2025, published in January 2025 ^[2]. It projects that 170 million jobs will be created and 92 million displaced between 2025 and 2030, a net gain of 78 million ^[2]. It puts total job disruption over the period at 22% of jobs ^[2].

These are not 'AI jobs'. The WEF's drivers include technology, demographics, geoeconomic tensions and economic pressures, and AI is only one of them ^[2].

What employers say they will do

• 77% plan to upskill their workers ^[2].
• 41% plan to reduce their workforce as AI automates tasks ^[2].
• Almost half expect to move staff from AI-exposed roles into other parts of the business ^[2].
• Half plan to reorient their business toward AI opportunities ^[2].

The reskilling gap

The WEF says nearly 40% of the skills needed on the job are set to change by 2030 ^[2]. 63% of employers name the skills gap as the main barrier to transforming their business ^[2]. If the global workforce were 100 people, 59 would need reskilling or upskilling by 2030 and 11 of them would be unlikely to get it ^[2]. That equals more than 120 million workers at medium-term risk of redundancy ^[2].

Why these are forecasts, not facts

The 2025 report is the fifth edition of the series. It draws on data from more than 1,000 companies in 22 industries and 55 economies ^[2]. That makes it a survey of employer expectations, not a measure of observed employment ^[2]. In October 2026 Georgieva wrote that it is still too early to predict AI's toll on employment ^[1]. Jed Kolko at the Peterson Institute for International Economics (PIIE) describes the evidence on AI's current labour-market effects as inconclusive ^[7].

Is AI already taking jobs? What the 2025–2026 data show

Using US data through mid-2026, the main US studies find no broad job loss from AI ^[3][4], and PIIE judges the overall evidence inconclusive ^[7]. Where researchers differ is on whether specific groups, especially young workers, are already being affected ^[3][7].

Stanford: no broad displacement, but a gap for young workers

The Stanford Digital Economy Lab published the August 2026 revision of its 'Canaries in the Coal Mine?' study, which uses ADP payroll data through June 2026 ^[3]. It finds no widespread, economy-wide job displacement associated with AI ^[3]. Employment of 22–25-year-olds in highly AI-exposed occupations is about 19% below where it would be had it kept pace with less-exposed peers, up from a 15% gap at the July 2025 data vintage ^[3]. Experienced workers show no comparable gap ^[3].

Between November 2022 and June 2026, employment of 22–25-year-olds in the two most AI-exposed quintiles fell about 11%. The same age group in the three least-exposed quintiles grew about 10% ^[3]. The adjustment works mainly through less hiring of young workers rather than layoffs ^[3]. So far it shows up in employment, not in base pay ^[3].

Yale: no clear evidence of disruption

The Yale Budget Lab's AI Labor Market Tracker was updated on 15 September 2026 with August 2026 Current Population Survey microdata ^[4]. It finds no clear evidence of labour-market disruption associated with AI ^[4]. Occupational churn, AI exposure among the unemployed and usage data are all flat, within historical ranges or on pre-AI trends ^[4]. In a 2026 review of what is and is not known, Yale Budget Lab researchers likewise found no clear evidence yet of AI effects on the labour market, but noted this could change quickly ^[8]. Writing for Brookings, Yale Budget Lab researchers found that the share of US workers in high-, medium- and low-exposure jobs stayed steady in the 33 months after ChatGPT launched in November 2022 ^[9].

IMF: no aggregate impact yet, but regional signals

In October 2026 Georgieva wrote that there is so far no apparent impact on the job market, but the nature of work is changing and this is likely to accelerate ^[1]. IMF research on Asia-Pacific, cited by Georgieva in October 2026, found that employer demand fell about 10% in highly AI-exposed occupations, even as demand grew for workers who can use AI ^[1]. A January 2026 IMF analysis found that in regions with high demand for AI skills, employment in AI-vulnerable occupations was 3.6% lower after five years than in regions with less demand ^[6].

Why the studies seem to disagree

• Different data. Stanford's estimates are larger in its ADP payroll sample than in national survey benchmarks ^[3].
• Correlation, not causation. The Stanford authors say their patterns are descriptive, not causal. The gaps shrink when education is accounted for, and some trends predate generative AI ^[3].
• Different exposure measures. PIIE notes that AI-exposure measures do not fully agree. It also cites an Eckhardt and Goldschlag (2025) study which found that unemployment rose less in high-exposure occupations ^[7].
• Low adoption. Citing the US Census Bureau's Business Trends and Outlook Survey, PIIE notes that fewer than one-fifth of US firms use AI in any capacity ^[7].
• AI-washing. PIIE warns that companies have an incentive to over-attribute hiring freezes or layoffs to AI ^[7]. Fortune's coverage of the Yale findings raised the same concern ^[10].

Which jobs is AI most likely to replace?

The evidence does not offer a reliable list of jobs AI will erase outright. Instead, it identifies roles with the most exposed tasks and roles employers expect to shrink ^[2][5]. Four patterns stand out.

Clerical and administrative work

The ILO finds that women are far more likely than men to hold jobs in the highest generative-AI exposure category, largely because women are overrepresented in clerical work ^[5]. The WEF lists cashiers and administrative assistants among the fastest-declining roles to 2030 ^[2].

Some creative production work

In its 2025 edition the WEF added graphic designers to the fastest-declining roles, linking the change to generative AI reshaping the labour market ^[2].

Routine middle-skill office jobs

The IMF's January 2026 analysis finds that high-skill and low-skill workers tend to gain most from demand for new skills ^[6]. Middle-skill roles, such as routine office jobs, are being squeezed ^[6].

Jobs where AI automates rather than complements

Automation is AI doing a task instead of a person. Augmentation is AI helping a person do the task better or faster. Stanford finds early employment declines concentrated in occupations where AI automates tasks, while employment is flat or rising where AI complements workers ^[3]. Employment fell among young workers in occupations that rely on codified knowledge, which is formal, documented knowledge ^[3].

Replacement is not the only risk

The ILO–NASK authors warn that partial automation can lower job quality even when the job survives. The risks they name are fewer responsibilities, stagnating wages and growing insecurity ^[5]. Because they see transformation as the most likely impact of generative AI, this erosion is a realistic concern for many workers ^[5].

Which jobs can AI not replace easily?

No job is guaranteed to be immune. Still, some kinds of work are much harder to automate with today's AI ^[2][3][11].

Frontline, care and education roles

The WEF expects frontline roles, namely farmworkers, delivery drivers and construction workers, to grow most in absolute terms by 2030 ^[2]. Next come care roles such as nursing professionals and education roles such as secondary teachers ^[2]. These projections reflect all the forces the WEF considers, including demographics, not just AI ^[2].

Work built on tacit knowledge

Tacit knowledge is know-how gained through practice and mentorship that is hard to write down. Stanford finds that employment rose among experienced workers in occupations that rely on tacit knowledge ^[3]. Over the same period it fell among young workers in codified-knowledge occupations ^[3].

Physical tasks in messy environments

According to UNU-C3's summary of the Stanford AI Index 2026, AI agents improved on structured computer-use tasks (the OSWorld benchmark) from roughly 12% to 66.3% accuracy ^[11]. Robots, however, still fail nearly nine in ten real household tasks ^[11]. On that evidence, screen-based work is open to AI far sooner than physical work in unpredictable settings.

Human skills that stay central

The WEF expects AI and big data, networks and cybersecurity to be the fastest-growing skills ^[2]. It names analytical thinking, resilience, leadership and collaboration among the critical core skills, and highlights creative thinking, flexibility and agility as human skills that are growing in importance ^[2]. New roles are appearing too: the IMF names prompt engineering and building AI agents as examples ^[1].

Will AI take my job? How to judge your own risk

This section gives general information, not personal career advice. For decisions such as changing field or paying for retraining, speak to a qualified careers adviser who knows your local job market.

Five questions to ask about your job

1. How much of my work is codified and screen-based? Young workers in occupations built on formal, documented knowledge have seen employment fall, and clerical work carries the highest generative-AI exposure ^[3][5].
2. Would AI do my tasks, or help me do them? Stanford's declines are concentrated where AI automates tasks. Where AI complements workers, employment is flat or rising ^[3].
3. Am I early in my career? The adjustment so far comes mainly through less hiring of young workers rather than layoffs ^[3]. The IMF also cites emerging US evidence that generative AI adoption reduces entry-level hiring ^[6].
4. How far has AI reached my workplace? Fewer than one-fifth of US firms use AI in any capacity, according to Census Bureau survey data cited by PIIE ^[7]. Pew Research Center surveyed 5,010 US workers on 2–8 September 2025. 21% said at least some of their work is done with AI, up from 16% about a year earlier, while 65% said they don't use AI much or at all ^[12].
5. Can I add a skill employers pay for? The IMF finds that about one in 10 job postings in advanced economies and one in 20 in emerging market economies require at least one new skill, and IT accounts for more than half of that demand ^[6]. In the UK and US, postings with a new skill pay about 3% more. Postings with four or more new skills pay up to 15% more in the UK and 8.5% more in the US ^[6].

Example: a recent graduate in an exposed office role

Picture a 23-year-old applying for junior roles that mostly involve drafting standard documents and summarising information. This is the group where Stanford finds the widest gap ^[3]. Employment of 22–25-year-olds in the two most exposed quintiles fell about 11% between November 2022 and June 2026, while the same age group in the least-exposed quintiles grew about 10% ^[3]. The practical signal is fewer openings rather than layoffs ^[3]. It makes sense to learn AI tools well, because demand is growing for workers who can use AI ^[1]. It also makes sense to keep practising the underlying skill, since heavy reliance on AI tools may slow skill development ^[11].

Example: a mid-career administrative assistant

Administrative assistants are among the WEF's fastest-declining roles ^[2]. Yet almost half of employers expect to move staff from AI-exposed roles into other parts of the business, and 77% plan to upskill workers ^[2]. Asking an employer about internal moves and training is a realistic first step.

Example: an experienced nurse or site manager

Nursing professionals and construction workers are among the roles the WEF expects to grow most by 2030 ^[2]. Experienced workers in tacit-knowledge occupations have seen employment rise ^[3]. For them, the likelier change is new tools inside the job rather than losing it.

What are the pros and cons of AI for society?

AI's effects reach beyond jobs into growth, inequality, energy use and fairness. Credible sources see real gains and real risks ^[1][7].

Benefits

• Faster growth. The IMF says AI's productivity boost could add about half a percentage point a year to global growth over the next decade ^[1].
• Productivity at work. Most firm-level studies reviewed by PIIE, many of them on coding and customer service, find productivity benefits from AI ^[7].
• New roles and pay premiums. New jobs such as prompt engineering and building AI agents are emerging ^[1]. Job postings that ask for new skills pay more in the UK and US ^[6].
• Rapidly improving tools. AI agents' accuracy on structured computer-use tasks rose from roughly 12% to 66.3%, according to the AI Index 2026 summary ^[11].

Drawbacks

• Inequality and public finances. The IMF warns that AI could reduce labour income, increase inequality and erode the tax base ^[1].
• Uneven productivity. In one study AI made developers slower. In another, consultants who over-relied on AI performed worse ^[7].
• Energy demand. Citing the International Energy Agency, the IMF says data centres are expected to account for 3% of global electricity consumption by 2030; the IMF adds that this is more than Japan consumes today ^[1].
• Bias in decisions. Algorithmic discrimination is unfair treatment produced by an automated system. The ILO highlights recruiting cases where biased training data led to unfavourable outcomes for female candidates ^[13].
• Skill erosion. The AI Index 2026 flags concern that heavy reliance on AI tools may slow skill development ^[11].
• Lower job quality. Partial automation can mean fewer responsibilities, stagnating wages and growing insecurity ^[5].

The IMF's growth figure describes a possible boost, not a guaranteed one ^[1]. PIIE stresses that the productivity evidence, while mostly positive, includes clear exceptions ^[7].

Who is most affected: young workers, women and poorer countries?

AI's effects are not spread evenly. The evidence points to four groups worth watching ^{[1][3][5][12]}.

Young and early-career workers

Stanford's employment gap for 22–25-year-olds in highly exposed occupations widened from 15% at the July 2025 data vintage to about 19% as of June 2026 ^[3]. Experienced workers show no comparable gap ^[3]. The IMF cites emerging US evidence that generative AI adoption reduces entry-level hiring ^[6]. PIIE, however, considers claims of harm to particular groups premature ^[7].

Women

Worldwide, 4.7% of women's jobs are in the ILO's highest generative-AI exposure category, compared with 2.4% of men's ^[5][13]. In high-income countries the figure is 9.6% of female employment, nearly three times the share for men, largely because women are overrepresented in clerical work ^[5][13].

Richer and poorer countries

The ILO finds that 34% of jobs in high-income countries are in generative-AI-exposed occupations, compared with 11% in low-income countries ^[5]. It cautions that lower exposure does not mean lower risk where digital access is limited or labour protections are weak ^[5]. On the IMF's AI Preparedness Index, advanced economies are three times better equipped in digital infrastructure than low-income countries ^[1]. IMF research likewise puts exposure higher in advanced economies, at about 60%, than worldwide, at roughly 40% ^[1].

Workers by education

In Pew's September 2025 survey, AI use at work among US workers with a bachelor's degree or more had risen from 20% to 28%, compared with a rise from 13% to 16% among those with some college or less ^[12].

How workers feel

In Pew's February 2025 report, based on an October 2024 survey, only 6% of US workers said workplace AI use will lead to more job opportunities for them in the long run ^[14].

What is different for readers in Iran?

Most of the evidence in this guide comes from the United States and other advanced economies, so readers in Iran should apply it with care ^[3][4].

Access to leading AI tools

Iran was not on OpenAI's list of supported countries ^[15]. OpenAI said it would enforce stricter blocking of API traffic from unsupported countries, including Iran, China, Russia and North Korea, from 9 July 2024 ^[15]. That means common advice to 'learn the leading AI tools' assumes access that many Iranian workers and developers do not formally have. Policies change, so check OpenAI's official supported-countries page for the current position. This guide does not cover ways around such restrictions.

Why global numbers may not fit

IMF research puts exposure at about 60% of jobs in advanced economies, against roughly 40% worldwide ^[1]. New-skill demand appears in about one in 20 job postings in emerging market economies, compared with one in 10 in advanced economies ^[6]. Stanford's ADP payroll findings and Yale's Current Population Survey findings describe the US labour market and do not transfer directly to Iran ^[3][4].

Low exposure is not low risk

The ILO warns that lower AI exposure does not mean lower risk where digital access is limited or labour protections are weak ^[5]. The IMF reports a threefold gap in digital infrastructure between advanced and low-income economies ^[1]. Both points matter for anyone whose internet access is filtered or disrupted, although none of the studies cited here measures Iran directly.

What is missing

There are no credible, published Iran-specific statistics on AI's labour-market effects among the sources reviewed for this guide, so any precise local estimate should be treated with suspicion. The skills the WEF calls core, such as analytical thinking, resilience, leadership and collaboration, do not depend on any one tool or platform ^[2].

What comes next, and what is still uncertain?

The evidence is still early. PIIE's Jed Kolko titled his March 2026 review 'Research on AI and the labor market is still in the first inning', and he judges current evidence inconclusive ^[7].

What is likely next

• Work keeps changing. The IMF says the nature of work is changing and this is likely to accelerate ^[1].
• Entry-level pressure is the signal to watch. Stanford's young-worker gap widened from 15% to 19% between data vintages ^[3].
• Skill demands shift. Nearly 40% of the skills needed on the job are set to change by 2030, according to the WEF ^[2].
• Infrastructure grows. Data centres are expected to reach 3% of global electricity consumption by 2030 ^[1].

What is still uncertain

• Cause and effect. Stanford's patterns are descriptive, not causal, and some trends predate generative AI ^[3].
• Speed. Yale Budget Lab researchers found no clear evidence of AI effects on the labour market in 2026, but noted this could change quickly ^[8].
• Adoption versus capability. Fewer than one-fifth of US firms use AI in any capacity ^[7], yet AI agents' performance on computer-use tasks has risen sharply ^[11].
• Who gains. The IMF warns that AI could reduce labour income and increase inequality, so the distribution of gains is an open question ^[1].

How to keep track

Two US trackers are updated with new data: Stanford Digital Economy Lab's 'Canaries in the Coal Mine?' revisions ^[3] and the Yale Budget Lab's AI Labor Market Tracker ^[4]. For a global employer view, the WEF's Future of Jobs series, whose 2025 report was the fifth edition, draws on data from more than 1,000 companies in 55 economies ^[2]. Until the data settle, the safest working assumption is that AI will reshape tasks first, and that the earliest effects show up in who gets hired rather than who gets fired ^[3].