AI and jobs in America: New data reveals who is truly at risk, and who isn’t

Over years, the artificial intelligence vs. jobs debate has been more about melodramatic forecasts: we will have a mass layoff, one day thousands of jobs will vanish, and workers will be forced to scramble for robotic-style jobs in order to earn their living. However, new reports published on March 5 by AI company Anthropic provide a more grounded vision, the one that implies that the actual effect of AI on jobs is much more unequal and complex than the narrative of doomsday.

Drawing on detailed usage data from its AI system Claude, the company introduced a new metric known as “observed exposure.” Unlike earlier projections that speculated about what AI might be capable of doing, this measure tracks how AI is actually being used inside real workplaces.

The results reveal a striking gap between AI’s theoretical abilities and its real-world adoption, a gap that continues to shield millions of jobs from immediate disruption.

Measuring AI in the real economy

To build what it calls the Anthropic Economic Index, the company’s researchers combined three major sources of data as reported by TheStreet.

First was the extensive occupational database maintained by the O*NET, which catalogues detailed task descriptions for nearly 800 professions across the United States.

Second were anonymised usage logs from Claude itself, offering a real-time snapshot of how organisations are deploying generative AI in daily workflows.

Finally, the study incorporated an academic framework developed in 2023 that evaluates whether AI could reduce the time required to complete a given task by at least 50 percent.

By merging these datasets, researchers assigned each occupation a “coverage score.” A high score indicates that AI is already performing a meaningful share of that job’s tasks in practice, while a score of zero means AI activity has yet to show up in real-world usage data. The results complicate many popular assumptions.

Even though AI systems could theoretically perform up to 90 percent of the tasks associated with office and administrative roles, the observed data tells a different story. In reality, AI usage currently covers only about one-third of tasks in the most affected category, computer and mathematics occupations.

Jobs already feeling the pressure from AI

The data analysed by Anthropic suggests that disruption is not evenly distributed across the workforce. Instead, it is concentrated in roles where tasks revolve around structured information, writing, coding, and digital communication, areas where large language models such as Claude excel.

Among the occupations showing the highest levels of observed AI exposure are:

Computer programmers and software developers, where AI tools are increasingly generating code, debugging systems, and automating documentation.

Customer service representatives, whose interactions are rapidly being replaced by automated chat systems and AI-driven support pipelines.

Data analysts and research assistants, whose work often involves summarising reports, extracting patterns from datasets, or drafting analytical text.

Technical writers and content creators, where AI can produce drafts, documentation, and explanatory material at scale.

Administrative and office employees, the monotonous documentation and scheduling, and reporting functions of which may be increasingly automated.

Scientists have highlighted that the functions are highly dependent on digital text processing and formalized workflows, which are specifically prone to AI augmentation or automation.

Employment projections from the Bureau of Labor Statistics reinforce this pattern, showing that occupations with higher AI exposure scores tend to have slightly weaker projected growth through 2034.

The professions AI still struggles to replace

In contrast, roughly 30 percent of the workforce shows no meaningful AI exposure in the observed data. These occupations tend to depend on physical presence, real-world judgement, and interpersonal awareness—capabilities that language models cannot replicate.


Examples include:

Electricians, plumbers, and construction workers, whose jobs require hands-on work in unpredictable physical environments.

Healthcare professionals such as nurses and paramedics, where physical care, patient interaction, and rapid decision-making remain central.

Teachers and classroom educators, who rely on emotional intelligence, real-time engagement, and adaptive instruction.

The employment estimates provided by the Bureau of Labor Statistics support this trend, with occupations with a higher AI exposure score reported to have a somewhat lower projected growth up to 2034.

Physical systems technologists like mechanics and equipment technicians who engage in diagnosing physical systems and not analysing text-based data.

In hospitality and service industries, such as chefs, hotels, and event organizers where human interaction and situational judgement are all that count.

These careers require a level of motor skills, spatial perception, and subtle human interaction, which the existing AI solutions do not have much practical access to.
Ready to navigate global policies? Secure your overseas future. Get expert guidance now!