--- title: "Which Jobs Will AI Eliminate? What the Unemployment Debate Is Missing" description: "Artificial intelligence will eliminate some tasks, reduce the size of some occupations, and redesign most jobs. However, high exposure to AI within an occupation does not mean that the same proportion of people working in that occupation will become unemployed." url: https://oihs.ca/en/blog/which-jobs-will-ai-eliminate lang: en date: 2026-08-26T19:29:33.189Z category: blog --- # Which Jobs Will AI Eliminate? What the Unemployment Debate Is Missing Artificial intelligence will eliminate some tasks, reduce the size of some occupations, and redesign most jobs. However, high exposure to AI within an occupation does not mean that the same proportion of people working in that occupation will become unemployed. ## Which Jobs Will AI Eliminate? What the Unemployment Debate Is Missing [Anthropic’s research published in March 2026 examines](https://www.anthropic.com/research/labor-market-impacts) the extent to which artificial intelligence may affect tasks across different occupations. The high exposure rates in the report were often presented by mainstream media as direct estimates of “unemployment caused by AI.” This led to scenarios predicting unemployment rates as high as 30–40%. In my view, these interpretations have also influenced university choices in Türkiye. I have seen young people and families become hesitant about computer engineering because of concerns that “AI will take programmers’ jobs.” Yet what the report actually says and what these headlines suggest are not the same thing. ## **High Exposure Does Not Mean High Unemployment** Anthropic’s theoretical capability measure looks at whether a task could be completed at least twice as quickly with the help of a large language model. According to this measure, 94% of tasks in computer and mathematical occupations could theoretically be accelerated. However, in the same report, observed exposure based on Claude’s actual workplace usage is only 33%. In other words, there is a substantial gap between what is technically possible and what is currently happening in the economy. More importantly, the study does not find a systematic increase in unemployment since late 2022 among occupations with greater exposure to AI. It does, however, identify an early signal—one that researchers emphasize should still be interpreted cautiously—that entry into these occupations has slowed among people aged 22–25. The conclusion we should draw from the report, therefore, is not “high exposure = high unemployment,” but rather: “The tasks within jobs are changing rapidly, and the entry point into the labour market for young people may be narrowing.” ## Occupational Fields Most Likely to Be Affected by AI ## **Occupational Fields Most Likely to Be Affected by AI** The table below presents the 22 broad occupational groups included in Anthropic’s March 5, 2026 report. “Theoretical task capability” refers to the potential for a large language model to complete tasks at least twice as quickly, while “observed exposure” refers to the share of professional tasks sufficiently represented in Claude usage during August and November 2025. Fully automated usage is given full weight, while usage in which AI supports a human is weighted at half. ## Occupational Fields Most Likely to Be Affected by AI | Rank | Occupational Field | Theoretical Task Capability | Observed Exposure | | --- | --- | --- | --- | | 1 | Business and finance | 94% | 29% | | 2 | Computer and mathematics | 94% | 33% | | 3 | Management | 91% | 14% | | 4 | Office and administrative support | 90% | 34% | | 5 | Legal | 87% | 20% | | 6 | Architecture and engineering | 84% | 4% | | 7 | Arts and media | 82% | 18% | | 8 | Life and social sciences | 76% | 4% | | 9 | Sales | 61% | 27% | | 10 | Education and library occupations | 60% | 18% | | 11 | Community and social services | 50% | 4% | | 12 | Protective services | 30% | 2% | | 13 | Healthcare support | 28% | 2% | | 14 | Production | 19% | 1% | | 15 | Installation, maintenance, and repair | 19% | 1% | | 16 | Personal care and services | 18% | 1% | | 17 | Construction | 17% | 1% | | 18 | Farming | 16% | 2% | | 19 | Food preparation and serving | 16% | 1% | | 20 | Transportation and material moving | 13% | 1% | | 21 | Building and grounds cleaning and maintenance | 13% | 1% | | 22 | Healthcare practitioners and technical occupations | 6% | 4% | ## Source and scope note: ***Source and scope note: Anthropic, “Labor market impacts of AI: A new measure and early evidence,” Figure 2, March 5, 2026. Percentages are based on the U.S. O\*NET task classification and Claude usage. They are not estimates of employment levels in Türkiye, job-loss rates, or the probability that an occupation will disappear.*** ## How Should We Read This Table? ## **How Should We Read This Table?** The fields with the highest observed exposure are office and administrative support (34%), computer and mathematics (33%), business and finance (29%), sales (27%), and legal occupations (20%). These fields largely consist of tasks carried out in digital environments and involving substantial amounts of text, data, coding, research, and communication. What is more striking than the ranking itself is the gap between theoretical capability and observed exposure. The difference is 80 percentage points in architecture and engineering, 77 points in management, and 72 points in life and social sciences. This gap illustrates the legal, technical, organizational, and human barriers between “AI can do this” and “organizations are currently having AI do this.” But this gap should not be treated as a safety margin. It may narrow as AI capabilities improve and organizational adoption increases. ## Tasks, Occupations, and Unemployment Are Not the Same Thing ## **Tasks, Occupations, and Unemployment Are Not the Same Thing** An engineer’s job does not consist only of making calculations, writing technical documents, or producing drawings. Engineers meet clients, visit project sites, choose between alternatives, assume risks, manage teams, and ultimately take responsibility for the work that is delivered. Artificial intelligence may accelerate some of these tasks dramatically. But the fact that one task can be automated does not mean the entire occupation can be automated. ### **There Is a Difference Between Theoretical Capability and Real-World Use** Even when it is technically possible to accelerate a task with artificial intelligence, organizations still need to install the software, prepare their data, train employees, and redesign workflows. Security, privacy, regulation, error risk, human verification, organizational culture, and resistance to change also play a role. Technological capability is not an economic outcome. It is merely one of the upper limits on what that economic outcome could become. ### **Productivity Gains Do Not Necessarily Cause Job Losses** If an engineer can complete the same amount of work in half the time with AI, a company could decide to employ one person instead of two. But another outcome is equally possible: engineering services become cheaper and faster, leading to greater demand. The company takes on more projects, and projects that were previously uneconomical become feasible. Computers automated much of the calculation work once performed by accountants. CAD software accelerated technical drawing. Excel transformed financial calculations. None of these occupations simply disappeared. Instead, the content of the jobs changed. ### **The Amount of Work in the Economy Is Not Fixed** The idea that “if people work twice as fast, half the workforce will become unemployed” assumes that the total amount of work available in the economy is fixed. But when the cost of a service falls, demand generally increases. If software development becomes cheaper, more software may be produced. If legal analysis becomes less expensive, more analysis may be requested. Without knowing how much of a productivity gain will translate into job losses and how much will translate into greater production, it is impossible to calculate an unemployment rate. ### **Job Loss and Unemployment Are Not the Same Thing** A person may lose their current job but move into another one shortly afterwards. Every year, millions of jobs disappear in the economy while millions of new ones are created. Therefore, the question “How many jobs will disappear?” is fundamentally different from the question “How much will the unemployment rate increase?” Comparing exposure, displacement, and persistent unemployment figures as though they represented the same thing is a serious category error. ## Why Lists of Jobs That Will Disappear in the Future Are Misleading ## **Why Lists of Jobs That Will Disappear in the Future Are Misleading** Search results are full of articles with titles such as “20 Jobs AI Will Destroy.” These lists often take task exposure from a research paper and turn it directly into an obituary for an entire occupation. Yet very different workflows can exist under the same job title. Technology may reduce prices and increase demand. Regulation and accountability may keep humans involved. The disappearance of one task may increase the importance of another. For this reason, it is more useful to look at a portfolio of tasks rather than the name of the occupation itself. Data entry, first-draft writing, standardized reporting, basic code generation, and routine customer responses may be more rapidly automated. Problem definition, verification, relationship management, implementation in the physical world, and ultimate responsibility may remain with humans for considerably longer. If two people hold the same occupation but one performs only routine tasks while the other manages clients, decisions, and responsibility, their levels of risk are not the sam ## What Does Research Say About AI and Unemployment? ## **What Does Research Say About AI and Unemployment?** * [IMF](https://www.imf.org/en/blogs/articles/2024/01/14/ai-will-transform-the-global-economy-lets-make-sure-it-benefits-humanity): Estimates that approximately 40% of global employment and around 60% of employment in advanced economies could be affected by artificial intelligence. Part of this exposure reflects substitution, while another part reflects the augmentation of human labour and productivity gains. * [ILO:](https://www.ilo.org/publications/generative-ai-and-jobs-2025-update) States that one in four jobs worldwide could be transformed by generative AI. Its central message focuses less on mass job destruction and more on the transformation of tasks and occupations. * [Goldman Sachs Research: ](https://www.goldmansachs.com/insights/articles/how-will-ai-affect-the-us-labor-market)Estimates that during a ten-year adoption period, 6–7% of U.S. workers may need to leave their current jobs, while the baseline scenario projects an increase of approximately 0.6 percentage points in the unemployment rate. * [KPMG](https://kpmg.com/kpmg-us/content/dam/kpmg/pdf/2025/gen-ai-economic-growth.pdf): In a rapid-adoption scenario combined with strong skills development, projects a net increase of 8.06 million jobs in the U.S. by 2050. In a rapid-adoption scenario without sufficient skills development, it estimates approximately 1 million net job losses and an additional 0.51 percentage-point increase in the unemployment rate. * [World Economic Forum](https://www.weforum.org/press/2025/01/future-of-jobs-report-2025-78-million-new-job-opportunities-by-2030-but-urgent-upskilling-needed-to-prepare-workforces/): Estimates that technological, economic, demographic, and environmental transformations—including AI—could displace 92 million jobs while creating 170 million new ones by 2030, resulting in a net gain of 78 million jobs. It is important to note that these figures do not measure the impact of artificial intelligence alone. These studies do not use the same methodology, country, or time horizon. For that reason, their figures should not be treated as though they belonged in a single league table. Nevertheless, their common message is clear: High exposure rates do not imply equally large levels of permanent unemployment. The final outcome depends not only on technology, but also on growth, investment, job creation, education, and transition policies. ## Why Lists of Jobs That Will Disappear in the Future Are Misleading ## **Why Lists of Jobs That Will Disappear in the Future Are Misleading** Search results are full of articles with titles such as “20 Jobs AI Will Destroy.” These lists often take task exposure from a research paper and turn it directly into an obituary for an entire occupation. Yet very different workflows can exist under the same job title. Technology may reduce prices and increase demand. Regulation and accountability may keep humans involved. The disappearance of one task may increase the importance of another. For this reason, it is more useful to look at a portfolio of tasks rather than the name of the occupation itself. Data entry, first-draft writing, standardized reporting, basic code generation, and routine customer responses may be more rapidly automated. Problem definition, verification, relationship management, implementation in the physical world, and ultimate responsibility may remain with humans for considerably longer. If two people hold the same occupation but one performs only routine tasks while the other manages clients, decisions, and responsibility, their levels of risk are not the same. ## What Does Research Say About AI and Unemployment? ## **What Does Research Say About AI and Unemployment?** * [IMF:](https://www.imf.org/en/blogs/articles/2024/01/14/ai-will-transform-the-global-economy-lets-make-sure-it-benefits-humanity) Estimates that approximately 40% of global employment and around 60% of employment in advanced economies could be affected by artificial intelligence. Part of this exposure reflects substitution, while another part reflects the augmentation of human labour and productivity gains. * [ILO:](https://www.ilo.org/publications/generative-ai-and-jobs-2025-update) States that one in four jobs worldwide could be transformed by generative AI. Its central message focuses less on mass job destruction and more on the transformation of tasks and occupations. * [Goldman Sachs Research:](https://www.goldmansachs.com/insights/articles/how-will-ai-affect-the-us-labor-market) Estimates that during a ten-year adoption period, 6–7% of U.S. workers may need to leave their current jobs, while the baseline scenario projects an increase of approximately 0.6 percentage points in the unemployment rate. * [KPMG:](https://kpmg.com/kpmg-us/content/dam/kpmg/pdf/2025/gen-ai-economic-growth.pdf) In a rapid-adoption scenario combined with strong skills development, projects a net increase of 8.06 million jobs in the U.S. by 2050. In a rapid-adoption scenario without sufficient skills development, it estimates approximately 1 million net job losses and an additional 0.51 percentage-point increase in the unemployment rate. World Economic Forum: Estimates that technological, economic, demographic, and environmental transformations—including AI—could displace 92 million jobs while creating 170 million new ones by 2030, resulting in a net gain of 78 million jobs. It is important to note that these figures do not measure the impact of artificial intelligence alone. These studies do not use the same methodology, country, or time horizon. For that reason, their figures should not be treated as though they belonged in a single league table. Nevertheless, their common message is clear: High exposure rates do not imply equally large levels of permanent unemployment. The final outcome depends not only on technology, but also on growth, investment, job creation, education, and transition policies. ## Will AI Increase Unemployment? ## **Will AI Increase Unemployment?** Probably—in some periods and in some occupations. Workers whose tasks are repetitive, clearly defined by rules, carried out digitally, and easily measurable are likely to face greater pressure. However, the evidence available today does not support a permanent and economy-wide unemployment scenario of 30–40%. In my view, the central issue is not the overall unemployment rate, but rather who will be affected by this transformation, how quickly, and with what bargaining power. If productivity gains go only to company owners while wages and working hours remain unchanged, social tension may increase. If, on the other hand, those gains are translated into new investment, shorter working hours, education, and new jobs, the same technology could produce a more inclusive outcome. ## The Real Risk: A Narrowing Entry Point for New Graduates ## **The Real Risk: A Narrowing Entry Point for New Graduates** The first visible effect of AI on employment may not be mass layoffs, but slower hiring. If a company dismisses 20 out of 100 employees, we see it immediately. But if a company that previously hired 20 new graduates every year now hires only five, nobody has technically been “laid off.” Yet the entry point into working life for young people has still become narrower. Anthropic’s research finds an approximately 14% decline, compared with 2022, in the rate at which 22–25-year-olds start new jobs in occupations with high AI exposure. The researchers explicitly emphasize the limitations of this finding, but it is nevertheless an important signal to monitor in the context of youth unemployment. The IMF’s 2026 assessment also points out that entry-level jobs have particularly high exposure to AI. There is a second consequence to this trend. If companies retain only experienced employees while shrinking entry-level roles, they may weaken the career ladder that develops the experienced workers of the future. The issue is therefore not only whether today’s young people can find jobs, but also whether organizations can sustain their future talent pipeline. ## Are There Jobs That AI Cannot Eliminate? ## **Are There Jobs That AI Cannot Eliminate?** I do not believe it is useful to provide a supposedly safe list of occupations that will never change. Even if an occupation does not disappear entirely, a significant proportion of its tasks may change. For now, however, jobs that involve physical environments, decision-making under uncertain conditions, trust and accountability, human relationships, or defining the problem itself appear more resilient. ### **What Jobs Can’t AI Do?** * Jobs requiring manual dexterity, physical movement, and adaptation to the physical environment * Roles in which legal, ethical, or commercial responsibility must ultimately be assumed by a human * Jobs requiring empathy, persuasion, negotiation, care, leadership, and trust-based relationships * Context-rich jobs in which the objective and the correct problem are not predefined * Roles that combine domain expertise across multiple disciplines But none of these should be interpreted as meaning “AI will not affect these jobs.” A more accurate statement would be this: In some occupations, AI will primarily increase human capability; in others, fewer people may be required to produce the same level of output. ## Will AI Create New Job Opportunities? ## **Will AI Create New Job Opportunities?** Yes, but we should not think only in terms of futuristic-sounding new job titles. The greatest opportunities may emerge in roles that combine existing domain expertise with artificial intelligence: clinical AI applications in healthcare, model risk management in finance, data and compliance in law, learning design in education, robotic systems integration in manufacturing, and data governance, cybersecurity, and AI quality assurance across virtually every sector. New jobs will not automatically be good or accessible jobs. If educational institutions fail to update their programs, companies do not provide entry-level workers with opportunities to learn, and employees cannot reskill, then the people losing jobs may not be able to transition into the new opportunities being created. For this reason, the question “How many new occupations will emerge?” is just as important as the question “How quickly will people be able to move into those jobs?” ## How Should Young People Prepare for the AI Era? ## **How Should Young People Prepare for the AI Era?** Simply telling young people “Don’t be afraid of AI” is not enough. Some entry-level tasks may genuinely decline. The right response is not to try to defend work that AI can already do, but to add AI’s capabilities to our own. * **Make AI part of your daily way of working.** Do not use it merely as a chatbot for asking questions. Use it for research, analysis, writing, coding, data processing, preparing presentations, and solving problems. * **Develop domain expertise**. A person who understands engineering, finance, law, education, or biology deeply and knows how to apply AI within that field will be more valuable than someone who “only knows AI.” * **Strengthen your verification skills.** AI can produce convincing but incorrect answers. Source checking, questioning assumptions, and interpreting data are becoming core professional competencies. * **Learn how to define problems.** As the cost of generating solutions decreases, the value of identifying which problems are actually worth solving increases. * **Invest in human relationships.** Persuasion, negotiation, leadership, teamwork, empathy, and trust-building are far more difficult to automate. * **Make your productivity gains visible.** Employers will increasingly ask not “Does this person use AI?” but “How much more value does this person create because they use AI?” Demonstrate this through projects, portfolios, and concrete outputs. * **Do not get trapped in one narrow task.** Workers who combine multiple skills are likely to be more resilient. * **Treat continuous learning as part of the job.** The shelf life of a diploma is becoming shorter. Regularly learning new tools and methods is no longer a separate activity; it is becoming part of working life itself. This approach is the new-technology version of the debate around 21st-century skills that I have emphasized for years. Technical knowledge matters, but by itself it is not enough without critical thinking, communication, collaboration, creativity, and learning how to learn. ## Frequently Asked Questions About AI and Jobs ## **Frequently Asked Questions About AI and Jobs** ### **Which Jobs Will AI Eliminate?** Current data does not provide a reliable list of occupations that will definitely disappear completely. The fastest changes are likely to occur in jobs with a high concentration of tasks involving text, data, coding, research, standardized reporting, and routine communication. What matters more than the occupation itself is the combination of tasks within it. ### **Will AI Increase Unemployment?** It may increase unemployment in some sectors and during transition periods. However, exposure rates are not the same as unemployment rates. Productivity, demand, job creation, wages, education, and social policies will determine the net outcome. ### **Are There Any Jobs That Will Not Be Affected by AI?** It is not safe to assume that any occupation will remain completely unaffected. Tasks involving physical execution, empathy, trust, decision-making under uncertainty, and ultimate responsibility appear more resilient, but the way these occupations are performed will also change. ### **How Should AI Risk Be Evaluated When Choosing a Career?** Do not look only at the job title. Consider how much of the daily work is routine and digital, the growth potential of the field, the degree of human interaction and responsibility involved, and the opportunities to develop skills that complement AI. Interest, aptitude, and willingness to learn remain among the strongest long-term indicators. ## Conclusion: Let’s Not Look for the Right Answer to the Wrong Question ## **Conclusion: Let’s Not Look for the Right Answer to the Wrong Question** “Which jobs will AI eliminate?” is an understandable question, but it often traps us in the wrong framework. More useful questions are: Which tasks will become faster? Which types of work will experience greater demand? How can young people’s entry into the labour market be protected? Who will benefit from productivity gains? How quickly can the education system prepare people for new tasks? Artificial intelligence will affect every occupation to some degree. It will eliminate some tasks, create some new jobs, and redesign most professions. I believe that in the future, the competition will not be between humans and artificial intelligence, but **between people who know how to use AI effectively and those who do not.**