Which Professions Can Artificial Intelligence Replace: Dario Amodei’s Forecast and Anthropic Data

Which Professions Can Artificial Intelligence Replace: Dario Amodei’s Forecast and Anthropic Data

While some professions are merely testing new tools, others are already facing the initial signs of a future restructuring of the labor market.

Artificial intelligence is taking over routine digital tasks fastest, and Anthropic CEO Dario Amodei predicts a severe blow to entry-level office jobs within the next 1–5 years. Data from Anthropic, the OECD, and the ILO show that physical work, contextual judgment, and high human responsibility currently remain less vulnerable.

What Dario Amodei Predicts About AI and the Labor Market

Photo: Dario Amodei at TechCrunch Disrupt 2023 in San Francisco, September 20, 2023. Source: TechCrunch / Wikimedia Commons, CC BY 2.0.

Dario Amodei is the co-founder and CEO of Anthropic, the company behind Claude. In October 2024, he published an essay titled “Machines of Loving Grace” on the potential consequences of very powerful AI. The section on work and the meaning of life begins not with a list of “jobs of the future,” but with a candid admission: economic outcomes are harder to predict, and the author does not have definitive answers.

In the near term, Amodei relies on the logic of comparative advantage. If AI performs 90% of the tasks in a given job better, the remaining 10% can make humans more productive and create new complementary roles. He separately notes that the physical world is likely to remain an area of relative or even absolute human advantage for a “considerable amount of time.” However, he offers no long-term guarantees: if AI becomes sufficiently versatile and inexpensive, he believes the current model of economic organization may cease to function.

In January 2026, the tone grew sharper. In the essay “The Adolescence of Technology”, Amodei reiterated his public 2025 estimate: AI could displace up to half of entry-level white-collar jobs within 1–5 years. At the same time, he explicitly clarified that he was not claiming this mass displacement was already occurring right now.

Among examples of areas with similar cognitive profiles, he cites finance, consulting, and law. His argument is that general-purpose AI could simultaneously impact multiple related fields, meaning that simply retraining from one office profession to another might prove less effective than during previous technological transitions.

Physical labor is also not a “safe haven” in this updated version of the forecast. Amodei notes that it may buy humans extra time, but robotics, autonomous vehicle operation, and AI-accelerated robot design could gradually erode this advantage as well. Likewise, he does not view the “human touch” as a sufficient universal remedy across the entire labor market.

Which Professions Are Most Vulnerable to Artificial Intelligence

Photo: construction workers at a site in New York, October 4, 2025. Source: Tessa Bury / Wikimedia Commons, CC BY 4.0.

The most constructive approach here is to distinguish the Anthropic CEO’s forecast from empirical measurements of what is actually happening today. In March 2026, Anthropic researchers published the observed exposure metric — the share of tasks within an occupation where Claude usage data already demonstrates real AI application, taking into account whether it involves full automation or human assistance. This is neither a measure of “layoff probability” nor a comprehensive assessment of the broader AI market.

In this sample, computer programmers showed the highest observed exposure — 75% of tasks covered. For data entry keyers, the figure stood at 67%, with customer support specialists also among the most exposed. At the other end of the spectrum, 30% of workers had zero observed exposure under the study’s threshold. Anthropic cited cooks, motorcycle mechanics, lifeguards, bartenders, dishwashers, and fitting room attendants among the examples.

“Zero exposure” does not mean these jobs cannot be automated. Rather, it indicates that the corresponding tasks appeared too infrequently in professional Claude usage to cross the study’s minimum threshold. The authors themselves point to tree trimming, operating agricultural equipment, and courtroom client representation as examples of uncovered tasks.

Even more importantly, high exposure does not currently translate into mass unemployment. In Current Population Survey data, Anthropic researchers found no statistically significant rise in unemployment among the most exposed occupations following the release of ChatGPT. For workers aged 22–25, they observed a weaker signal: the likelihood of starting a new role in a highly exposed occupation was approximately 14% lower than in 2022, though this result was borderline statistically significant and open to alternative explanations.

An independent perspective from the OECD provides a similar yet broader framework. Their 2026 index compares current AI capabilities with occupational requirements across nine cognitive, social, and physical domains. Today, AI comes closest to roles involving routine data processing, administrative procedures, and well-codified tasks. The greatest distance remains where contextual judgment, understanding human behavior, complex decision-making, and accountability are required.

In its 2025 global index, the International Labour Organization analyzed nearly 30,000 tasks. According to its estimates, one in four workers globally is employed in an occupation with some degree of exposure to generative AI. However, because human involvement remains necessary, the ILO regards job transformation as a far more likely scenario than the outright disappearance of most occupations.

Consequently, it is more accurate to analyze task characteristics rather than compiling a list of “surviving” professions. In the short term, jobs set in unpredictable physical environments, as well as roles requiring personal accountability, contextual judgment, trust, and nuanced human interaction, appear more resilient. Conversely, digital tasks that can be clearly documented, replicated, verified, and scaled algorithmically remain far more vulnerable.

Moreover, this boundary is constantly shifting. Amodei explicitly cautions that advances in robotics could diminish the current advantage held by physical labor, while the OECD emphasizes that the real impact of AI depends not only on technological capabilities, but also on adoption rates, regulation, labor organization, and societal decisions. A low-exposure profession today carries no permanent guarantee, just as a high-exposure profession is not doomed to extinction.

The practical takeaway from these findings is simpler than any “top jobs of the future” ranking: evaluate the specific bundle of tasks within an occupation. The more a role demands ownership of results, interaction with the physical world, navigating ambiguous scenarios, interpersonal trust, and human coordination, the harder it is to automate completely. The more repetitive digital routine it involves, the faster AI may absorb individual functions — even if the job title itself endures.