Quick Read
- Anthropic’s new study finds AI could theoretically automate 94% of computer tasks, but observed usage is only 33%.
- AI exposure falls hardest on higher-paid, college-educated workers, not low-wage service workers.
- Computer programmers, customer service reps, and data entry specialists are among the most exposed professions.
- No significant increase in unemployment has been detected among AI-exposed occupations since late 2022.
- Hiring for young workers (22-25) in AI-exposed fields dropped 14% since 2024, signaling a possible structural shift.
WASHINGTON (Azat TV) – A new study by AI research company Anthropic, released this week, reveals that artificial intelligence is poised to impact high-paid, college-educated workers most significantly, overturning previous assumptions that low-wage service roles would bear the brunt of automation. The study introduces a critical distinction between AI’s theoretical capabilities and its actual deployment in the workplace, indicating a substantial gap that currently limits widespread job displacement.
The research, which combines occupational data, theoretical AI exposure scores, and real-world usage metrics from Anthropic’s Claude AI, found that while AI could theoretically speed up 94 percent of all computer and math tasks, observed Claude usage covers only 33 percent of these. This “theory vs. reality” gap underscores that the full disruptive potential of AI has not yet materialized in practical workplace applications, despite its advanced capabilities.
AI’s Observed vs. Theoretical Impact on Tasks
Anthropic’s methodology, detailed in their labor market study, developed a metric called “observed exposure.” This metric carefully measures how much of AI’s theoretical potential has translated into actual use in work contexts. It assigns tasks a value based on whether a language model alone can double the speed of unassisted work, or if additional tools are required, or if there’s no meaningful speed advantage. This conservative measure focuses on work-related contexts and heavily weights fully automated API use, providing a floor estimate of AI’s influence.
The study contrasts this observed exposure with a theoretical baseline derived from the ‘GPTs are GPTs’ paper by Eloundou et al. (2023), which assessed the potential reach of large language models across hundreds of U.S. occupations. The stark divergence shows that while 97 percent of tasks performed by Claude are theoretically feasible for AI, a significant portion of automatable tasks are simply not being handled by the AI in practice. For instance, tasks like authorizing medication refills, scored as fully automatable, have not been observed in actual Claude usage.
White-Collar Roles Face Unexpected Exposure
Contrary to popular narratives, Anthropic’s findings indicate that AI exposure falls hardest on demographics typically associated with higher education and better pay. The study found that workers highly exposed to AI hold jobs paying 47 percent more on average than their unexposed counterparts, with nearly four times as many college graduates in the exposed group. These workers are also more often female and white, a demographic profile that inverts the conventional expectation of AI primarily threatening manual or service labor.
The study identifies computer programmers as topping the list with 74.5 percent observed task coverage, followed by customer service representatives (70.1 percent), and data entry specialists (67.1 percent). Other highly exposed roles include legal assistants, technical writers, financial analysts, medical record specialists, market research analysts, sales representatives, software quality assurance analysts, and information security analysts, according to CBS News. These are predominantly knowledge-work occupations involving text processing, analysis, and code generation. Conversely, jobs requiring significant physical ability and embodied judgment, such as cooks, motorcycle mechanics, lifeguards, and bartenders, show little to no observed AI coverage.
Early Warning Signals for Young Workers
Despite high theoretical exposure in some professions, employment data has not yet revealed a significant increase in unemployment among AI-exposed workers since ChatGPT’s release in late 2022. The Anthropic researchers stated that their methodology is sensitive enough to detect a doubling of the unemployment rate in exposed occupations, which has not occurred.
However, a specific segment of the workforce does show an early warning signal: hiring for workers aged 22 to 25 in AI-exposed occupations has declined since 2024. The job-finding rate for this group dropped by roughly half a percentage point, amounting to a 14 percent decline over the post-ChatGPT period. No comparable decline was observed for workers over age 25. This suggests that while experienced workers in AI-exposed fields may be insulated for now, entry-level positions could be absorbing the first wave of AI-driven efficiency gains, with companies potentially hiring fewer junior staff while expecting the same output from existing headcount. The study frames this as a possible leading indicator of structural change, rather than confirmed displacement.
Broader Research Corroborates AI’s Complex Impact
Anthropic’s findings align with independent research from several other teams, lending further credibility to the nuanced picture of AI’s impact. A 2025 Danish study of 25,000 workers, for example, found no measurable changes in wages or working hours despite documented high AI usage. Similarly, a Microsoft Copilot study of 200,000 conversations identified knowledge workers as AI-exposed, while cautioning against equating AI capability with full automation – a central argument Anthropic also advances.
MIT economist David Autor’s Stanford study, reviewing occupational data from 1977 to 2018, concluded that automation outcomes depend on whether routine tasks are removed alongside expert ones. Dallas Fed research further distinguishes between codified knowledge (which AI handles well) and tacit knowledge rooted in experience (which resists automation). Across these studies, a consistent theme emerges: AI adoption is shaping long-run labor market projections and compressing entry-level hiring, but has not yet generated the widespread unemployment spikes that would confirm structural displacement, as reported by WinBuzzer.
The Anthropic study’s ‘observed exposure’ metric serves as a crucial monitoring instrument, offering a reality check against alarmist predictions by highlighting the persistent gap between AI’s theoretical potential and its actual integration into the labor market. This distinction provides policymakers and employers with a more flexible response timeline, focusing attention on managing potential structural shifts in entry-level hiring rather than immediate mass unemployment.