Artificial intelligence has generated no shortage of discussion about its benefits and, perhaps even more so, its potential risks. In the workplace, one concern stands above the rest: if AI can perform our work faster and better than we can, will it eventually eliminate our jobs?
The concern is understandable. A tool that can process documents, summarize information, generate code, answer questions, and perform complex calculations in seconds can complete certain tasks thousands of times faster than a person.
But performing some of a person’s tasks is not the same as replacing everything that person does.
A job is not a single activity. It is a collection of tasks. Some are digital, repetitive, document-based, or relatively easy to standardize. Others depend on contextual knowledge, experience, accountability, negotiation, exception handling, validation, and decision-making under uncertainty.
This leads to an important distinction when assessing AI’s impact on work: automating part of the work does not necessarily mean automating the worker.
We have seen comparable transformations before
Earlier industrial revolutions displaced particular tasks and occupations, but they also produced new roles and areas of expertise.
The introduction of semi-automated machinery, control systems, sensors, industrial robots, and software reduced a great deal of manual and repetitive work. At the same time, it increased demand for machine operators, maintenance technicians, programmers, automation specialists, systems integrators, and engineers capable of implementing and supervising these technologies.
That does not mean every displaced worker moved seamlessly into one of the new roles, nor does it prove that every technological transition created more jobs than it eliminated. Such transitions can produce unemployment, inequality, and the obsolescence of once-valuable skills.
What they do show is that technology rarely operates as a simple one-for-one replacement of people. It also changes the tasks within occupations and the capabilities employers expect from the people performing them.
AI may be driving a comparable shift. This time, however, the frontier of automation is moving beyond primarily physical and mechanical work into cognitive activities that, until recently, were widely regarded as uniquely human.
AI exposure does not equal job elimination
To explore this relationship, I examined data from ILOSTAT alongside research from the International Labour Organization (ILO), the OECD, and other specialist sources.
In 2025, the ILO and Poland’s NASK research institute published an updated methodology that analyzed nearly 30,000 tasks and classified occupations by their degree of exposure to generative AI.
The study found that approximately one in four workers worldwide is employed in an occupation with some degree of GenAI exposure. Yet only about 3.3% of global employment falls within the highest exposure category.
The differences across income groups are also substantial:
- High-income countries: approximately 34% of employment has some degree of exposure.
- Worldwide: approximately 24%.
- Low-income countries: approximately 11%.
This may appear counterintuitive. More developed economies have greater exposure because GenAI currently affects digital, cognitive, administrative, analytical, and document-based work most directly. By comparison, automating many physical activities still requires perception, movement, object manipulation, safety systems, maintenance, infrastructure, and significant capital investment.
The current wave of automation is therefore not concentrated solely among less-skilled industrial workers. It also reaches administrative and professional occupations that once appeared comparatively protected from technological displacement.
There is, however, a crucial distinction to preserve. These indicators measure the potential exposure of occupational tasks. They do not measure actual use of ChatGPT or similar tools, and they do not count how many workers have been replaced by them.
The ILO explicitly warns against treating exposure indexes, on their own, as forecasts of job losses. High exposure means that a substantial share of an occupation’s tasks could be performed or supported by GenAI. It does not mean that the entire job will necessarily disappear.
Exposure is not distributed evenly
When I explored the available data for Chile, I found 500,066 women and 338,413 men in the GenAI exposure category under examination.
Those absolute figures do not, by themselves, prove that an individual woman is more likely to be exposed. Establishing that would require comparing each figure with total female and male employment. What the result did show was a larger number of women in the occupational group being examined, raising a question worth investigating.
One possible explanation was the higher concentration of women in administrative, clerical, accounting, secretarial, and document-intensive occupations—the very roles that contain large numbers of digital, text-based, repetitive, and standardizable tasks.
The ILO’s international findings point in the same direction.
Within the highest exposure category worldwide, the organization estimates that exposure affects:
- 4.7% of female employment;
- 2.4% of male employment.
In high-income countries, the gap is wider:
- Women: 9.6%.
- Men: 3.5%.
This does not mean women are more exposed simply because they are women. It primarily reflects how women and men are distributed across occupations with different task profiles.
The Chilean result, then, does not appear to be an isolated anomaly. It is consistent with a broader international occupational pattern.
There was an important limitation in the Chilean data, however. In some of the annual GenAI exposure datasets, Chile appeared only in 2022. Those data cannot support a genuine trend analysis or establish whether exposure increased or decreased in subsequent years.
Tasks may be a more useful unit of analysis than jobs
Even highly exposed occupations continue to include activities that require human involvement.
An AI system can draft a document, but someone must determine whether it fits the relevant context. It can generate code, but someone must validate its functionality, security, and integration. It can recommend a response to a customer, but human intervention may still be necessary when exceptions, sensitive consequences, negotiation, or accountability are involved.
For that reason, the first thing to lose ground may not be the job itself, but some of the tasks people currently perform manually.
This can free time for more valuable work, but it can also reduce the number of people required to produce the same output. If a company once needed ten employees and six AI-assisted workers can now handle the same workload, the transition does not necessarily begin with four immediate layoffs.
It may instead take the form of:
- reduced hiring;
- natural attrition without replacement;
- internal reassignment;
- changing job profiles;
- higher productivity expectations;
- new responsibilities for existing employees.
Looking only at announced layoffs may therefore cause us to miss a significant part of AI’s effect on labor demand.
The earliest signal may be appearing among younger workers
Recent evidence points to a particularly important development.
The Stanford Digital Economy Lab has been analyzing millions of U.S. administrative payroll records provided by ADP. Its August 2026 update, based on data through June, found no evidence of widespread employment displacement across the economy as a whole.
Once the researchers separated workers by age and occupational exposure, however, a substantial difference emerged.
Among workers aged 22 to 25 in highly AI-exposed occupations, employment stood approximately 19% below where it would have been had it kept pace with employment among young workers in less-exposed occupations.
Using the same comparison, the gap was about 15% in the July 2025 data and had widened to 19% by June 2026.
Looking at the raw change between November 2022 and June 2026:
- employment among workers aged 22 to 25 in the two most-exposed quintiles fell by approximately 11%;
- employment for the same age group in the three least-exposed quintiles grew by about 10%.
This does not mean that AI causally destroyed 19% of these jobs. The researchers describe the result as a descriptive divergence and acknowledge that other economic and sector-specific factors may be involved.
Even so, the pattern remains when they examine alternative explanations involving interest rates, remote work, technology companies, computer-related occupations, and firm entry and exit.
The mechanism is equally important: the divergence appears to be driven primarily by reduced hiring of younger workers, rather than by a wave of dismissals among people already employed.
Is AI replacing the economic function of junior workers first?
A separate study by Harvard researchers, presented at Stanford, examined data covering nearly 62 million workers and 285,000 U.S. firms between 2015 and 2025.
The researchers compared companies showing signs of GenAI adoption with companies that did not. Beginning in 2023, they observed a relative decline in junior employment at adopting firms, while senior employment continued to grow.
Once again, the main mechanism was slower hiring rather than a comparable increase in employee departures.
Both studies have limitations and cannot attribute the entire divergence exclusively to AI. Still, together they point toward a hypothesis that deserves greater attention: AI may be replacing part of the economic function of a junior worker before eliminating the position of an experienced one.
A senior professional brings relationships, experience, process knowledge, organizational context, and accumulated judgment. Someone at the beginning of a career generally spends a larger share of their time on initial research, classification, documentation, first drafts, basic analysis, and relatively standardized programming.
These are precisely the areas in which GenAI has made some of its greatest advances.
This creates an important paradox: if we automate the tasks through which junior workers traditionally gain experience, how will we develop the senior professionals of the future?
Jobs can evolve, but the transition will not happen automatically
The available evidence does not show that AI will eliminate every job. Nor does it establish that the technology will necessarily create more positions than it displaces.
So far, it also does not show an economy-wide employment collapse proportional to the degree of occupational exposure.
What is beginning to emerge is a reconfiguration of tasks, required skills, and entry points into certain careers. The impact may first appear through lower hiring, changing job profiles, or the contraction of particular functions, well before it appears as mass layoffs.
Something comparable happened with computers and search engines. Using these technologies effectively was once a distinguishing skill. Over time, it became part of the basic digital literacy expected across a wide range of occupations.
AI may follow a similar path. Instead of producing millions of people whose sole profession is writing prompts, it may lead to lawyers, accountants, doctors, analysts, developers, and professionals across many fields who know how to use, supervise, and validate AI systems within their own areas of expertise.
The ability to work with these tools may eventually cease to be an exceptional advantage and become a transferable skill as commonplace as using a computer or searching for information online.
A conclusion that remains open
Based on the evidence available through September 2026, my central hypothesis is this: the first observable labor-market effect of generative AI does not appear to be a massive decline in aggregate employment, but a transformation of tasks, the demand for experience, and the entry points into certain occupations.
This does not mean jobs are protected or that everyone will benefit equally from the transition. A decline in junior opportunities could become one of its most serious consequences, precisely because it threatens the process through which people acquire the experience required to perform more complex work later in their careers.
The question should no longer be limited to how many jobs AI can replace.
We should also ask which tasks it is absorbing, how hiring is changing, what skills will become necessary, and how we will build new learning pathways for people entering the workforce.
AI can replace work without necessarily replacing the worker. But if it changes the tasks through which people learn how to work, its impact may prove far deeper than the number of layoffs appearing in the headlines.
Primary sources
- International Labour Organization and NASK, Generative AI and Jobs: A Refined Global Index of Occupational Exposure (2025): view source.
- ILO, methodological guidance on occupational exposure indicators (2026): view source.
- Stanford Digital Economy Lab, Canaries in the Coal Mine?, August 2026 update: view source.
- Hosseini and Lichtinger, Generative AI as Seniority-Biased Technological Change: view source.