SI explained

Will SI take my job? What the studies actually measure

Empty office chair next to a briefcase whose right half dissolves into glowing amber pixels
Illustration created with Claude (Anthropic).

Headlines about SI (AI) and jobs quote exposure indices, employer surveys and payroll data as if they were the same thing. We explain what each study counts, what has changed so far and who is most exposed.

Will SI take my job, or in the older wording, will AI take my job? It is one of the most searched worries about SI (AI), and the headlines give answers that seem to contradict each other: "40% of jobs exposed", "92 million jobs displaced", "no discernible disruption". They are not all wrong. They measure different things. This guide explains what the main studies actually count, what has happened in the labour market so far, who is most exposed, and how to think about your own job without either panic or false comfort. It is not career advice and it does not predict any single person's future.

Three different questions hiding behind one

Most of the confusion comes from mixing three questions that researchers answer with different methods:

  1. Exposure: could SI do some of the tasks in this job, in principle? This is measured by comparing task lists with what models can do. It says nothing about whether employers will actually use SI, or whether they will cut staff if they do.
  2. Employer plans: what do companies say they intend to do? This is measured with surveys. It reflects intentions, which can change.
  3. Observed change: what has happened to employment and hiring since SI tools spread? This is measured with payroll and survey data. It only covers the past, and only a few years of it.

A large exposure figure and a small observed effect are not a contradiction. They describe a technology that touches many tasks but has, so far, changed jobs more slowly than the exposure numbers suggest.

The headline numbers, and what each one counts

StudyFigureWhat it actually measures
Eloundou, Manning, Mishkin and Rock (March 2023)About 80% of US workers could have at least 10% of tasks affected; about 19% at least halfTask exposure to large language models in the United States, with no timeline for adoption
IMF (January 2024)Almost 40% of global employment exposed; about 60% in advanced economiesExposure, split between jobs that may benefit and jobs that may lose demand
ILO Working Paper 140 (May 2025)One worker in four in an occupation with some exposure; 3.3% in the highest categoryGraded exposure of occupations worldwide
WEF Future of Jobs Report 2025 (January 2025)170 million jobs created and 92 million displaced by 2030, a net gain of 78 millionEmployer survey covering all drivers of change, not SI alone

Read the last column before the second. The paper by Tyna Eloundou and colleagues, first posted on arXiv on 17 March 2023, states plainly: "We do not make predictions about the development or adoption timeline". It also found that higher-income jobs may face greater exposure, which runs against the older idea that automation hits low-paid work first.

The International Monetary Fund's figures, presented by Kristalina Georgieva on 14 January 2024, are exposure figures too. In advanced economies, she wrote, about half of exposed jobs may benefit from SI through higher productivity, while for the other half SI could take over key tasks, which may lower labour demand, wages and hiring. For emerging markets and low-income countries the IMF put exposure at 40% and 26%.

The World Economic Forum figures are the ones most often quoted as "jobs lost to AI", and they are not that. The report, published on 8 January 2025 and based on more than 1,000 companies in 22 industries and 55 economies, forecasts job disruption equal to 22% of jobs by 2030 from technology, demographic shifts, geoeconomic tensions and economic pressures combined. SI is one driver among several.

Exposure is not job loss

A job is a bundle of tasks. If SI can draft the first version of a report, the person who wrote reports does not disappear; part of the bundle changes. The International Labour Organization's 2025 index, written by Pawel Gmyrek and colleagues and published on 20 May 2025, reaches that conclusion in so many words: "transformation of jobs is the most likely impact of GenAI", because most occupations still contain tasks that need people.

What an exposure study cannot tell you is what an employer does with the time saved. It can produce more with the same team, hire fewer new people, or cut existing staff. The first two are far more common in the early data than the third, as the next section shows.

The same applies to SI itself. Today's systems are fast and often useful, but they still state false things with confidence and do not judge their own reliability, which is why so much of the work around them is checking. Our guide to what SI is and what it can and cannot do covers those limits; the article on AGI, ASI and SI explains why "a system that can do any job" is still a hypothesis rather than a product.

What has actually happened so far

Two of the most cited studies of real employment data point in the same direction, with one important exception.

No economy-wide disruption yet

The Budget Lab at Yale, in an analysis by Martha Gimbel, Molly Kinder, Joshua Kendall and Maddie Lee published on 1 October 2025, compared how fast the mix of occupations in the United States changed in the 33 months after ChatGPT's release in November 2022 with earlier waves of computers and the internet. Their conclusion: "the broader labor market has not experienced a discernible disruption since ChatGPT's release 33 months ago". The mix of jobs was changing only about one percentage point faster than at the turn of the century, and that faster pace had started in 2021, before generative SI. The share of workers in the most and least exposed occupations stayed stable at around 18% and 29%.

The authors are careful about what this means. They note that past technologies took about a decade to transform office work, and that it is "too soon to tell" how disruptive SI will be. Stability so far is evidence about the past, not a promise.

The exception: young workers in exposed jobs

Erik Brynjolfsson, Bharat Chandar and Ruyu Chen at the Stanford Digital Economy Lab use payroll data from ADP covering millions of US workers. Their paper "Canaries in the Coal Mine?" first appeared in August 2025; the revised version of 12 August 2026 uses data through June 2026. It finds "no evidence of widespread, economy-wide job displacement", but one clear gap: employment of workers aged 22 to 25 in SI-exposed occupations "now stands 19% below where it would be had it kept pace with that of their less-exposed peers". Experienced workers show no comparable gap.

Three details matter more than the headline:

  • The gap comes mainly from fewer hires of young workers, not from people being fired.
  • It is concentrated in occupations where SI use mostly substitutes for human tasks. Where SI mostly complements workers, employment is flat or rising.
  • The authors call these "early, descriptive indicators" rather than causal estimates. The pattern weakens when controlling for education, some trends predate generative SI, and it is stronger in the ADP sample than in national surveys.

Put together, the evidence so far says: no wave of layoffs across the economy, but a narrowing entry door in some occupations. If you are early in your career in a field built on tasks SI does well, that is the signal to watch.

Who is most exposed

The studies agree more on who than on how much.

  • Clerical and administrative work. The ILO finds that clerical occupations "continue to have the highest exposure levels". The WEF lists cashiers and administrative assistants among the fastest declining roles.
  • Some creative and digital roles. The WEF report adds graphic designers to its list of declining jobs, and the ILO notes rising exposure in highly digitised professional and technical occupations.
  • Women, more than men. In the ILO index, 4.7% of women's employment falls in the highest exposure category against 2.4% of men's, because women are over-represented in clerical work. In high-income countries the gap is 9.6% against 3.5%.
  • Richer countries. Exposure is 34% of total employment in high-income countries and 11% in low-income countries, according to the ILO. The IMF reaches a similar ordering.
  • New entrants. The Stanford data point to young workers in exposed occupations, through slower hiring.

Jobs with physical, in-person or care components are the least exposed. The WEF expects the largest absolute growth by 2030 among farmworkers, delivery drivers, construction workers, nursing professionals and secondary school teachers.

What SI does to the people who keep their jobs

For many workers the more immediate change is not losing a job but doing it differently. The best-known field study, by Erik Brynjolfsson, Danielle Li and Lindsey Raymond and published in the Quarterly Journal of Economics in 2025, followed 5,179 customer support agents as an SI assistant was rolled out. Issues resolved per hour rose by 14% on average and by 34% for novice and low-skilled agents, with minimal effect for the most experienced and skilled.

That result cuts both ways. It suggests SI can spread the know-how of the best workers to newcomers, which helps people learn faster. It also suggests why employers might need fewer junior hires to reach the same output, which fits the Stanford pattern. Neither study shows that experienced workers become unnecessary.

Do new jobs appear?

History suggests they do, but not automatically and not for the same people. David Autor, Caroline Chin, Anna Salomons and Bryan Seegmiller traced job titles across eight decades of US census data. They found that the majority of current employment is in job specialties introduced after 1940, and that new work tends to appear where technology complements what an occupation produces. They also found a warning: the demand-eroding effect of automation has grown stronger over the last four decades, while the demand-raising effect of augmenting technologies has not.

The WEF report illustrates the transition problem. Its 170 million new jobs and 92 million displaced ones are not the same people in the same places. The report estimates that 59 workers in every 100 will need retraining by 2030, and that 11 of them are unlikely to receive it. Among surveyed employers, 77% plan to upskill staff, while 41% plan to reduce their workforce as SI automates certain tasks.

How to think about your own job

No study can tell you what will happen to you. You can, however, ask the questions the researchers ask:

  1. List your tasks, not your title. Which ones are mainly reading, writing, summarising or producing standard documents? Those are the most exposed. Which ones need judgement on unclear situations, physical presence, relationships or responsibility for a decision?
  2. Ask whether SI would replace or assist each task. The Stanford data separate the two: employment held up where SI complements workers and fell for young workers where it substitutes.
  3. Watch hiring, not just layoffs. The first signal in the data is fewer junior openings. If your field stops hiring beginners, the pipeline is changing even if nobody is fired.
  4. Learn the tools and their limits. The customer support study shows the largest gains for people who use the assistant well. Knowing when output is wrong, and checking it, is part of that skill. If your work involves reviewing text or images from others, a free SI detector can be one check among several, though no detector is a verdict on its own.
  5. Check where a number comes from. Before you share a scary or reassuring statistic, ask: is it exposure, an employer survey or observed data? Which country, which years, and does it include every cause of change or SI alone?

Reading the next headline

Expect more numbers. Exposure indices are updated as models improve, the Stanford team publishes a public set of indicators to track the data, and the Yale team says it will keep monitoring monthly. A useful habit is to put each new figure in one of the three boxes from the start of this guide. A forecast of jobs "affected" is not a count of jobs lost, and a quiet labour market today is not proof that nothing will change.

The same caution applies to the flood of content about SI itself, much of which is produced quickly and recycled without sources. Our pieces on SI slop and on the dead internet theory look at that side of the story. For the work question, the honest summary in October 2026 is short: SI is changing tasks in many jobs, it has not yet caused economy-wide job losses in the data, and the clearest early pressure is on people trying to get their first job in the most exposed fields.

FAQ

Will AI take my job?

Studies do not predict individual outcomes. Exposure research, such as the ILO 2025 index, finds that most jobs are more likely to be transformed than replaced, because they still contain tasks that need people. Observed data from Yale (October 2025) and Stanford (August 2026) show no economy-wide job displacement so far in the United States, but fewer hires of 22 to 25 year olds in the most exposed occupations.

How many jobs will AI replace?

No reliable count exists. The most quoted figure, 92 million jobs displaced by 2030 from the World Economic Forum Future of Jobs Report 2025, covers all drivers of change, not AI alone, and is set against 170 million jobs created. Exposure figures such as the IMF's almost 40% of global employment measure jobs where AI could affect tasks, not jobs that will disappear.

Which jobs are most at risk from AI?

Clerical and administrative occupations have the highest exposure in the ILO index, and the WEF lists cashiers, administrative assistants and graphic designers among fast-declining roles. Exposure is higher for women than for men and higher in rich countries than in poor ones. Jobs with physical, in-person or care work are the least exposed.

Has AI already caused job losses?

Not at the scale of the whole economy, according to the Yale Budget Lab (October 2025) and the Stanford Digital Economy Lab (revised August 2026). The Stanford paper does find that employment of workers aged 22 to 25 in AI-exposed occupations is 19% below where it would be had it kept pace with less-exposed peers, mainly through reduced hiring. The authors present this as an early descriptive signal, not a causal estimate.

Does AI make workers more productive?

In some settings, yes. A study of 5,179 customer support agents published in the Quarterly Journal of Economics in 2025 found that an AI assistant raised issues resolved per hour by 14% on average and by 34% for novice and low-skilled agents, with minimal effect for the most experienced workers.

Sources

  1. Generative AI and Jobs: A Refined Global Index of Occupational Exposure (ILO Working Paper 140) · International Labour Organization, 20 May 2025
  2. AI Will Transform the Global Economy. Let's Make Sure It Benefits Humanity · International Monetary Fund, 14 January 2024
  3. Future of Jobs Report 2025: 78 Million New Job Opportunities by 2030 but Urgent Upskilling Needed · World Economic Forum, 8 January 2025
  4. GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models · arXiv, 17 March 2023
  5. Evaluating the Impact of AI on the Labor Market: Current State of Affairs · The Budget Lab at Yale, 1 October 2025
  6. Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence (revised) · Stanford Digital Economy Lab, 12 August 2026
  7. Generative AI at Work · NBER / Quarterly Journal of Economics, 2025
  8. New Frontiers: The Origins and Content of New Work, 1940-2018 · NBER / Quarterly Journal of Economics, 2024

Spotted an error? Write to hello@siornot.com. Corrections are made in the article and the update date changes. Read our editorial policy.