How Will AI Affect Jobs?

Overview

Artificial intelligence (AI) is more likely to change many jobs by automating or assisting particular tasks than to eliminate every job in which it is used. Its employment effects can run in opposite directions: AI may reduce the number of workers needed for some tasks, help employees produce more or better work, increase demand by lowering costs, and create new work that requires human judgment or technical expertise. Which effect dominates depends on the occupation, how employers reorganize work, whether demand grows, and how the gains are distributed.

Current evidence does not support a precise prediction of total jobs AI will eliminate or create. The International Labour Organization (ILO) estimates that one in four workers worldwide is in an occupation with some exposure to generative AI, but only 3.3% of global employment is in its highest exposure category. The ILO judges job transformation more likely than wholesale redundancy because most occupations still include tasks requiring human input (Generative AI and Jobs: A Refined Global Index of Occupational Exposure | International Labour Organization). Exposure means that tasks could be affected; it is not a forecast that a worker will lose a job.

How AI changes jobs

Jobs are bundles of tasks

An occupation is not a single activity. A claims adjuster, accountant, teacher, software developer or nurse performs a mix of information gathering, routine processing, communication, judgment and responsibility. AI usually affects some parts of that mix before it changes the whole job. Predictive systems can classify, rank or estimate; generative systems can draft text, summarize information, create images or code, and respond to prompts. Robotics and autonomous systems can also affect physical tasks, but their capabilities and constraints differ from those of language or image-generation tools.

A useful economic framework separates three effects. Displacement occurs when a machine takes over a task previously performed by a worker. Augmentation occurs when AI makes a worker more productive at tasks that remain human-led. Reinstatement occurs when technology creates new tasks in which people have a comparative advantage, such as designing, maintaining, supervising or applying a system to a specific service. These mechanisms can occur at the same time; their net effect on hiring and wages is ambiguous (Automation and New Tasks: How Technology Displaces and Reinstates Labor; Artificial intelligence and jobs: No signs of slowing labour demand (yet): OECD Employment Outlook 2023 | OECD).

Diagram: From task capability to labor-market outcome. AI capability by itself does not decide whether a job disappears. Employer adoption, work redesign, demand for the output, and the availability of other tasks determine how task changes translate into employment.

A productivity gain does not mechanically produce a layoff. An employer may use faster production to handle more customers, improve service, reduce prices, or increase profits; it may also decide to deliver the same output with fewer staff. If demand for a product expands when its price falls, productivity can support hiring. If demand is fixed and the technology substitutes for labor, the employer may cut vacancies or staffing. The effects can differ between a company, an occupation, and the economy as a whole.

Exposure is not the same as replacement

Exposure indexes estimate the degree to which the tasks in an occupation overlap with what AI systems may be able to do. They do not directly measure whether an employer has adopted a tool, whether it works reliably in that setting, whether a human must review its output, or whether the firm plans to change staffing. A high exposure score can indicate substantial changes to the content of work without predicting that the occupation will disappear.

The ILO's 2025 index combined a representative sample of occupational tasks, survey responses from 1,640 employed people, expert discussion and AI-assisted predictions. Its analysis covered nearly 30,000 tasks in the occupational classification and collected 52,558 data points on perceived automation potential for 2,861 tasks. It places occupations in four exposure gradients. The authors report that clerical occupations remain most exposed, while some highly digitized professional and technical occupations have become more exposed as generative models improve (Generative AI and Jobs: A Refined Global Index of Occupational Exposure | International Labour Organization). These are structured estimates of task capability, not a count of jobs certain to be automated.

Different studies use different definitions. The IMF's 2024 analysis estimated that nearly 40% of global employment was exposed to AI, including about 60% in advanced economies, 40% in emerging-market economies and 26% in low-income countries. Its measure covers AI more broadly and is not directly comparable to the ILO's 2025 generative-AI gradients (AI Will Transform the Global Economy. Let’s Make Sure It Benefits Humanity.). Neither percentage means that the same share of jobs will be lost.

What evidence shows so far

Productivity can rise, but findings depend on the task

Controlled studies show that AI assistance can improve performance on bounded tasks. In a preregistered experiment, 453 college-educated professionals completed occupation-specific writing assignments with or without ChatGPT. Participants with access to the tool completed the tasks 40% faster on average, and independent evaluators rated their output quality 18% higher. The experiment measured selected writing tasks, not long-term workplace employment or productivity across an entire occupation (Experimental evidence on the productivity effects of generative artificial intelligence).

A workplace study followed 5,179 customer-support agents after a generative-AI conversational assistant was introduced. Access to the tool increased issues resolved per hour by 14% on average and by 34% for novice and lower-skilled agents, with minimal effects for the most experienced and highly skilled workers. The researchers suggest the system helped less-experienced agents use patterns of successful interactions. The study is a working paper and concerns one company's customer-support operation; its measured productivity increase does not establish a specific effect on headcount (NBER WORKING PAPER SERIES).

These results illustrate why the question “how does AI affect jobs?” often has a task-level answer first. A tool that drafts routine responses may help an employee handle more cases, reduce the need for some routine work, and shift the employee's time toward exceptions or difficult conversations. Which outcome follows depends on the employer's operating choices and the volume and type of work.

Workplace accounts include benefits and concerns

An OECD survey of 5,334 workers and 2,053 firms in finance and manufacturing across seven countries found that about 80% of workers who used AI said it had improved their performance, while 8% said it had worsened it. These were users' reported experiences, not an experimental estimate of productivity. In the same survey, workers reported concerns about job stability and wages; employers were more likely to report that AI had decreased employment than increased it, though most reported no overall employment change. The survey also found substantial task reorganization: 66% of finance employers and 72% of manufacturing employers using AI said it had automated tasks, while about half in each sector said it had created tasks not previously done by workers. Three-quarters of AI users said AI had increased the pace of their work, which can mean greater output but may also mean more intense work (The impact of AI on the workplace: Main findings from the OECD AI surveys of employers and workers (EN)).

The OECD findings are limited to two sectors and seven countries, and many results are self-reported. Earlier OECD reviews of employment studies found little statistically significant evidence of aggregate job losses from AI exposure before the recent wave of generative AI. The OECD cautioned that adoption had been limited and that older evidence might not describe what happens as large language models are incorporated into business processes (Artificial intelligence and jobs: No signs of slowing labour demand (yet): OECD Employment Outlook 2023 | OECD).

More recent evidence offers an early warning about entry-level hiring, but not a final causal verdict. A Stanford Digital Economy Lab working paper, revised in August 2026, analyzes ADP payroll data covering millions of U.S. workers through June 2026. It reports no evidence of widespread economy-wide displacement, but finds that employment among workers aged 22–25 in AI-exposed occupations is 19% below the level it would have reached if it had kept pace with less-exposed peers. The divergence is primarily associated with reduced hiring rather than increased separations. The authors explicitly describe the pattern as descriptive rather than causal; it attenuates when education is controlled for, some divergence predates generative AI, and the ADP sample may not represent all workers (Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence - Stanford Digital Economy Lab). It is a signal to monitor, not proof that AI alone caused the gap.

Taken together, the evidence does not show that productivity gains automatically become either job creation or job loss. It shows that task performance can improve, tasks and work pace can change, and hiring effects can be uneven. The translation from tool performance to labor-market outcomes takes time and depends on adoption, investment in complementary processes, employer demand, and the distribution of productivity gains.

Which jobs and workers are likely to be affected?

Clerical and information-processing work

Clerical and administrative occupations have high exposure because many tasks involve processing standardized information, maintaining records, scheduling, preparing routine documents or handling predictable queries. The ILO's 2023 analysis found 24% of tasks in clerical-support jobs had high exposure to automation and another 58% medium exposure, while emphasizing that partial automation could leave substantial human work in place (u ILO Brief). The ILO's refined 2025 analysis continues to identify clerical occupations as the most exposed.

The World Economic Forum's 2025 employer survey expects clerical and secretarial roles, including cashiers, ticket clerks, administrative assistants and executive secretaries, to decline by 2030. Respondents attributed these expected declines to multiple forces, including AI and information processing, digital access, robotics, slower growth and demographic change; the projections do not attribute every decline to AI alone (Future of Jobs 2025: Jobs Outlook | World Economic Forum).

Professional and technical occupations

AI exposure is not confined to routine or lower-paid work. Generative tools can draft or summarize documents, search and organize information, produce code, and assist with analysis. This puts some tasks in law, finance, media, design, software and research within reach of AI systems. In these settings, workers may spend less time producing a first draft or processing standard cases and more time checking results, handling exceptions, explaining decisions, coordinating work or taking responsibility for outcomes. That shift may reduce demand for some entry-level tasks even when senior roles continue to require people.

The WEF survey lists AI and machine-learning specialists, big-data specialists, fintech engineers, and software and applications developers among the roles employers expect to grow quickly. It also expects certain clerical roles to decline. These are employer expectations across several economic and technology trends, not a guarantee that every worker in a growing occupation will benefit or that AI alone will create the projected positions (Future of Jobs 2025: Jobs Outlook | World Economic Forum).

Care, education, trades and physical work

Jobs centered on physical presence, hands-on care, interpersonal trust, complex environments or responsibility for people may be less amenable to full automation by current generative AI. Their tasks can still be affected: AI may help with documentation, scheduling, translation, decision support or training. Robots and other technologies can change manual work, but that is distinct from the capabilities of text-generating systems.

The WEF expects growth in care roles, education, delivery, construction and farm work, driven by demographic, economic, technological and green-transition trends in combination. For example, its survey links expected growth in care jobs to aging populations, and growth in education roles to digital access and working-age population changes. These forecasts should not be read as evidence that AI itself will create those jobs (Future of Jobs 2025: Jobs Outlook | World Economic Forum).

Gender and national differences

Exposure follows existing employment patterns. The ILO's 2025 global estimates place 4.7% of women's employment and 2.4% of men's employment in the highest generative-AI exposure gradient. The difference is associated with women's greater representation in clerical and administrative occupations. The ILO also estimates overall exposure at 34% of employment in high-income countries and 11% in low-income countries; lower exposure in poorer economies does not necessarily mean they are better positioned to benefit, because access to infrastructure, skills and investment also matters (Generative AI and Jobs: A Refined Global Index of Occupational Exposure | International Labour Organization; AI Will Transform the Global Economy. Let’s Make Sure It Benefits Humanity.).

Chart: Share of women's and men's employment in the ILO's highest GenAI exposure category. The gender comparison reflects employment shares; it is not a count of workers forecast to be displaced. The separate income-group figures in the caption refer to any exposure gradient and should not be compared directly with the bars.

Job creation, job loss and forecasts

Employers can respond to AI by hiring fewer people for a task, reallocating existing staff, reducing vacancies through attrition, or expanding output and creating new roles. Those adjustments may take years. Employment totals can hide important changes within a firm: growth in technical or supervisory roles can occur alongside fewer clerical or junior positions. Wages and job quality may also change even where the number of jobs stays stable.

The World Economic Forum's Future of Jobs Report 2025 estimates that several macrotrends combined will create 170 million jobs and displace 92 million current jobs worldwide by 2030, for a net increase of 78 million. Those figures are not an AI-only forecast. For its specific category of AI and information-processing technologies, employers expect 11 million jobs created and 9 million displaced between 2025 and 2030 (Future of Jobs 2025: Jobs Outlook | World Economic Forum).

The WEF projections use a survey of 1,043 employers representing more than 14.1 million employees across 55 economies and 22 industry clusters. The survey was conducted in 2024 and primarily targeted organizations with at least 500 employees. WEF mapped respondents' expected workforce changes to ILO employment data; the occupations with sufficiently usable survey data covered 1.18 billion employees, a subset of the global workforce. Small enterprises and informal work were outside the survey's intended scope, and the report notes that its selected-workforce estimates are not comprehensive (Future of Jobs Report 2025 Methodology | World Economic Forum). The numbers are conditional projections based on employer expectations and a mapping methodology, not realized job counts.

Earlier labor-market evidence also needs careful interpretation. The OECD found no statistically significant aggregate employment decline in several studies of AI exposure before widespread generative-AI use, but noted adoption was low and firms might initially adjust through attrition rather than layoffs. By contrast, evidence of productivity improvement on selected tasks is already measurable. Neither kind of result settles the long-term net effect. U.S. Bureau of Labor Statistics projections similarly emphasize that long-term estimates are conditional descriptions based on assumptions, not precise predictions of the future; the agency says rapid AI development makes the employment effects especially uncertain (Artificial Intelligence (AI) impacts on employment projections : U.S. Bureau of Labor Statistics).

Skills, wages and working conditions

Most people working alongside AI will not need to build machine-learning models. They will need to understand how AI tools fit their work, when outputs require verification, and how to combine those tools with occupation-specific knowledge. An OECD analysis of online vacancies across 10 countries found that highly AI-exposed occupations commonly requested management, business-process, social and digital skills; the report says most workers exposed to AI will not require specialized AI-development skills (Artificial intelligence and the changing demand for skills in the labour market (EN)).

The likely skill changes differ by role. A customer-service worker may need to check AI-generated responses and handle unusual cases. A programmer may review generated code, test security and maintain system behavior. A professional using AI to summarize records must judge whether the summary omits relevant facts. Across these examples, verification, communication, domain knowledge, problem-solving and accountability remain important even when AI performs part of the production process.

Productivity gains do not guarantee higher wages. They may be shared with workers through pay, training, reduced routine burdens or shorter hours; alternatively, they may accrue mainly to employers or owners, particularly if workers have limited bargaining power or the technology substitutes for their labor. The IMF warns that differences in access to AI, skills and complementary work can widen inequality within and between countries. OECD survey results also show why job quality must be considered alongside job counts: users reported positive effects in many cases, but also higher work pace and concern about employment and wages.

Managing the transition

Employers can reduce avoidable harm by assessing tasks rather than assuming an entire occupation can be removed, testing tools for quality and failure modes, involving workers in work redesign, and tracking whether productivity improvements translate into better service, wages, working conditions or staffing decisions. The OECD survey found that worker or representative consultation was associated with more positive reported productivity and working-condition outcomes, though that association does not by itself prove consultation caused the improvement (The impact of AI on the workplace: Main findings from the OECD AI surveys of employers and workers (EN)).

For governments and education systems, relevant responses include accessible retraining, help with job transitions, effective safety nets, and monitoring of hiring, wages and job quality by age, gender, occupation and industry. Training should combine practical AI literacy with skills that fit the worker's occupation rather than presume that everyone must become an AI specialist. Policymakers also face a distribution question: who receives the productivity gains, and what support is available to people whose tasks or jobs are displaced? The OECD calls for education and training and social dialogue; the IMF highlights labor-market mobility, human capital and social protection as components of preparedness (Artificial intelligence and jobs: No signs of slowing labour demand (yet): OECD Employment Outlook 2023 | OECD; AI Will Transform the Global Economy. Let’s Make Sure It Benefits Humanity.).

Conclusion

AI will affect jobs mainly by changing the tasks workers perform, the skills employers seek, and the way firms decide how many people to hire. Some routine information-processing work is likely to shrink; many other jobs will be reorganized, and AI can raise productivity or create complementary work. Current global exposure estimates and employer forecasts identify pressure points but cannot establish how many jobs will ultimately disappear. Evidence to date supports a definite conclusion about the direction of change—more task redesign and uneven effects—while the scale of net job creation, displacement and wage change remains unsettled.