Which professions will artificial intelligence quickly replace by 90%: results of a new study
Analysts from Anthropic have gathered data on what share of work tasks artificial intelligence is already capable of taking on in various professions today.
Image: vecteezy
The American company Anthropic, which specializes in AI safety and development, has published a large-scale study on the impact of technologies on the labor market. The company, founded in 2021 by former OpenAI employees, is known for its family of large language models, Claude, which directly influence the transformation of modern work processes.
The authors introduce a fundamentally new indicator — observed exposure. Unlike previous studies that only estimated what AI could theoretically do, this method combines theoretical models with real data on how people use Claude for professional purposes (based on the Anthropic Economic Index). The report is based on the analysis of millions of tasks, correlated with the O*NET database, which describes about 800 unique professions in the USA.
The chasm between theory and practice
The study reveals a significant gap between the theoretical potential of AI and its actual application. The corresponding diagram presents two indicators (blue and red, respectively): the theoretical potential of AI (what part of tasks in a profession AI can, in principle, perform) and its actual use today.
Source: anthropic.com
The data shows that AI is still far from reaching its maximum capabilities. For example, in fields such as IT, business, finance, and administration, AI can theoretically accelerate the execution of 90-96% of tasks. However, in practice, in the same IT specialties, the actual coverage is currently only 33%.
This gap is explained by legal barriers, security requirements, the need for human verification, and the slow diffusion of technologies into the corporate sector.
Who is under attack?
The study confirms that AI affects completely different categories of workers than previous waves of automation. While industrial robots replaced physical labor, AI primarily affects "white-collar" workers.
The most vulnerable professions named are programmers, customer support staff, and financial analysts.
Top 10 professions most affected by artificial intelligence. Source: anthropic.com
On the opposite side of the spectrum are professions where AI is practically powerless. This is work that requires physical presence and complex motor skills: chefs, mechanics, lifeguards, bartenders.
At the same time, about 30% of all jobs in the US currently have zero AI coverage.
Socio-demographic portrait of a vulnerable worker
The authors found a clear pattern: workers in fields most susceptible to AI's impact differ significantly from others. They:
- have a high level of education: the proportion of people with academic degrees is four times higher here;
- earn on average 47% more;
- are more often women (16% more compared to non-vulnerable areas).
Has unemployment begun?
The authors compared their data with forecasts from the US Bureau of Labor Statistics for 2024–2034. As of early 2025, researchers found no evidence of a systemic increase in unemployment among vulnerable professions. The dynamics of layoffs in these sectors currently align with overall market trends.
However, a disturbing trend has been recorded concerning youth aged 22-25. Although those who already have jobs are not losing them, the ability for young specialists in AI-vulnerable fields to find new employment has noticeably decreased. Data shows a 14% drop in the hiring rate of young people into these professions after the release of ChatGPT.
This indicates that companies are beginning to use AI to perform basic, routine tasks traditionally assigned to "juniors" (beginners). Thus, graduates of higher and secondary specialized educational institutions are primarily affected, as the entry threshold into the profession becomes higher for them.
Forecasts and the future
The US Bureau of Labor Statistics is already beginning to account for these changes: professions with high observed vulnerability to AI have lower employment growth forecasts. According to calculations, every additional 10% coverage of tasks by artificial intelligence reduces the employment growth forecast for that profession by 0.6 percentage points.
Anthropic notes that this study is only the first step. The developed methodology allows for regular data updates and tracking how the "red zone" of real AI usage expands. This will help economists and government agencies to timely distinguish deep structural changes from temporary market fluctuations and prepare appropriate support measures for the workforce.