When a wall gives way, the first warning is almost never an immediate collapse. It’s usually a hairline crack in the paint that’s hardly noticeable until you’re looking for it. Something similar happens with AI and employment: three years of cataclysmic forecasts that outline a labor apocalypse are, for now, only a crack.
A team at the Barcelona Institute of Economics looked for that gap, occupation by occupation, over the past five years of employment data. The result allows us to draw some interesting conclusions about the impact of AI on those most exposed to it.
A small gap, but with direction. A working paper published by the researchers compared 502 occupations with SEPE data between January 2021 and March 2026. For every ten additional points in AI, unemployment for each occupation increases by 1.2% in the 25 months after ChatGPT is introduced at the end of 2022.
After that time, as AI became more present in companies, the unemployment rate increased by 1.7% for every ten points of exposure. However, the authors themselves admit that this data is imprecise and cannot be considered for statistical purposes, but rather as a reference indicator of a trend.
Code, data and design on the front lines. To measure job exposure to AI, the team of researchers used Anthropic’s measure, which analyzes the actual use of clades in each job. Programmers top the list of occupations most open to task automation with AI, with 75% of their jobs already covered. They are followed by data recorders, sales agents and financial analysts. Software and multimedia analysts and designers also make it into the top 10 occupations with the highest risk of losing their jobs to AI.
The map of occupations affected by AI drawn by the IEB researchers is very similar to that revealed by research data conducted by Anthropic where they did not find a significant increase in unemployment among the most exposed occupations, although they did find some indication of fewer junior hires. Researchers at the University of Barcelona talk about “accumulations of unemployment”, but not a reduction in the number of jobs in the occupations most associated with AI.
Employment is not going down, but it is not going up. The small cracks that the IEB researchers found in the Spanish labor market were not the only ones that went undetected. According to a report by BBVA Research and FedEA, employment in occupations exposed to AI has not only registered the job destruction that some had predicted, but its employment rate is rising above average, neutralizing the negative correlation between exposure and unemployment.
According to the data obtained by the survey, fewer people are employed in occupations exposed to AI, but nevertheless, the number of registered unemployed in these occupations is high. In other words, even if people are employed in these professions, unemployment is accumulating. The authors do not examine whether more professionals are joining the sector or whether the unemployed take longer to find a job.
Youth, again at the center. Stanford researchers found that employment in AI-exposed occupations among 22- to 25-year-olds is 19% lower than expected. According to its authors, this happens because fewer are hired, not more fired. That is, it is attributed that AI has changed the profile of employees needed by companies, giving preference to those with more experience.
According to data collected from Anthropic’s own workforce, junior profiles went from 27% to 9% of its technical hires between early 2025 and June 2026. In Spain, the IEB survey found that unemployment among workers aged 30 to 39 rose further, up to 3.4% in every ten categories. On the other hand, contracts in the most open occupations also increased for people under 30, although probably because of higher turnover and not because they had more stable jobs.
In Xataka | Tech firms cut 92,000 jobs to invest in AI. The problem is the layoffs are costing them a fortune.
Picture | Pexels (Cottonbro Studios)
