introduces an "observed exposure" metric, comparing theoretical AI capabilities with real-world usage data from their Claude model. This research suggests that while AI's theoretical reach is vast, actual adoption and displacement are currently a fraction of what's possible.
According to Anthropic's findings, large language models are theoretically capable of handling 94% of tasks in computer and math occupations and 90% in office and administrative roles. However, observed usage indicates Claude currently covers only 33% of computer and math tasks and a fraction of administrative tasks in professional settings. This gap highlights a significant potential for future disruption as AI capabilities advance and adoption deepens.
Anthropic CEO Dario Amodei has previously warned that AI could displace up to half of entry-level white-collar jobs within five years, potentially leading to an "unusually painful" shock in the labor market. Despite these warnings, Anthropic's economists, Maxim Massenkoff and Peter McCrory, note that there is "limited evidence that AI has affected employment to date," with no systematic rise in unemployment in highly exposed fields.
The research indicates that workers most at risk are often older, more educated, and higher-paid, including roles in business, finance, law, and computer science. Conversely, jobs requiring physical presence and dexterity, such as cooks, mechanics, and bartenders, show zero AI exposure. While widespread job losses haven't materialized, the study identifies early signs of a slowdown in hiring for young workers in AI-exposed occupations.
