The impact of artificial intelligence on employment has become one of the defining debates of our time, with international organizations, academics, and employment companies regularly publishing predictions about which occupations are most at risk.
advertisement
advertisement
A new entry into that crowded field is being made by one of the AI giants.
Anthropic, the company behind Claude, has released a report titled AI’s Labor Market Impact: New Measures and Early Evidence, based on proprietary real-world usage data.
Theoretical Capacity and Observed Exposure
This report introduces a new measure called “observed exposure.” This measure is designed to quantify not only which tasks large-scale language models can theoretically speed up, but also which tasks are already automated in practice.
The distinction is important. Theoretical capabilities reflect what the AI can do, and observed exposures reflect what the AI actually does.
The highest scope of theoretical AI: computers, mathematics, business, finance.
Theoretical AI coverage exceeds 80% in several of the 22 occupational groups analyzed. Computers and Mathematics and Business and Finance occupations have the highest theoretical AI coverage, both at 94.3%.
Other groups with more than 80% theoretical competency include management (91.3%), office and administrative support (90%), legal (89%), architecture and engineering (84.8%), and arts and media (83.7%).
Five additional occupational groups have LLM penetration greater than 50%.
These include life and social sciences (77%), sales (62%), education and library professions (61.7%), health care workers (59.9%), and social services (50.5%).
The red area above shows how people are using Claude in professional settings, based on data from the Human Economy Index.
“As capabilities advance, adoption broadens, and deployment deepens, the red area will grow to cover the blue. There is also significant uncovered space. Of course, many tasks remain beyond the reach of AI, from physical agricultural tasks such as pruning trees and operating farm machinery to legal tasks such as representing customers in court,” the report said.
Lowest “likelihood” includes transportation, agriculture, and food.
The lowest theoretical AI adoption rate is in the ground maintenance sector, with only 3.9% of jobs in this group theoretically enabling the use of AI.
Transportation (12.1%), Agriculture (15.7%), Food and Catering (16.9%), Construction (16.9%), Personal Care (18.2%), Installation and Repair (18.4%), and Production (19%) also have significantly lower theoretical AI coverage, all below 20%.
This suggests that there may be less potential scope for using AI in these areas.
Theoretical AI coverage is also lower in medical support (28.5%) and protective services (31.6%).
Highest Exposures Observed: Computers, Mathematics, Offices, Administrators
A more important question is to what extent theoretical capabilities translate into observed exposures, indicating AI displacement risk.
Computer and mathematics occupations have the highest observed AI adoption rate at 35.8%, closely followed by office and management occupations (34.3%).
Business and Finance (28.4%) and Sales (26.9%) are also close to these levels.
Law-related occupations (20.4%), arts and media (19.2%), and education and library occupations (18.2%) also observe relatively high AI exposure at around 20%.
Observed exposure as a share of theoretical AI capabilities
The ratio of observed exposure to theoretical capacity indicates how much of this potential has already been used.
Sales topped the list at 43% (27% vs. 62%), followed by office and administrative jobs (38%) and computer and math jobs (38%).
The observed exposure as a share of theoretical AI capabilities is 30% for business and finance, education and library occupations.
Architecture and Engineering have a very high theoretical ability (85%), but the proportion is only 5%.
Occupations most at risk: Computer programmers and customer service representatives
Among individual occupations, computer programmers had the highest rate of exposure to AI at 74.5%.
Customer service representatives (70.1%), data entry keystrokers (67.1%), and medical records professionals (66.7%) also rank among those most at risk.
Market research analysts and marketing professionals followed at 64.8%, followed by sales representatives in wholesale and manufacturing industries, excluding technical and scientific products (62.8%).
This data also reveals who is most at risk. Workers in the highest-risk occupations tend to be older, better educated, better paid, and more likely to be women.
But the exposure has not led to job losses, at least so far.
The report found that unemployment rates have not increased systematically among workers in at-risk occupations since late 2022, but did find suggestive evidence that employment among young workers has slowed in the same sectors. This detail is worth noting.
