The most useful AI job chart right now is not the one that predicts mass layoffs. It is the one that shows how uneven the transition still is.
Anthropic’s latest labor-market research tries to measure something more grounded than abstract forecasts: not what large language models might eventually do, but where they are already appearing inside real work. Its new “observed exposure” framework combines task-level job data with actual Claude usage, weighting automated and work-related uses more heavily. That makes the results less dramatic than many viral claims about AI and more revealing about where pressure is building first.
At the top of the list are occupations built around structured digital tasks. Anthropic says computer programmers show 75% coverage, with customer service representatives and data entry keyers close behind. Medical record specialists, market research analysts, marketing specialists, financial and investment analysts, software quality assurance analysts, information security analysts, and computer user support specialists also rank high in exposure. The company defines exposure as the share of job tasks that AI can plausibly speed up or help perform, then narrows that estimate using real-world Claude usage.
That distinction matters. The study argues that AI remains far from its technical ceiling in the workplace. In Anthropic’s own example, just 33% of tasks in Computer & Math are currently covered, even though the theoretical potential is much higher. The gap points to a familiar pattern in technology adoption: capability arrives before organizations, workflows, and trust catch up.
So far, the measurable labor-market shock still looks limited. Anthropic found no systematic increase in unemployment for workers in highly exposed occupations since late 2022. It did, however, find evidence that deserves attention: hiring appears to have softened for younger workers entering exposed fields. In its analysis of workers ages 22 to 25, entry into highly exposed jobs fell by about 14% relative to 2022, a signal that AI may be affecting the front door of white-collar work before it affects the payroll rolls of established employees.
That helps explain why entry-level office work has become the center of the AI labor debate. Many of the tasks easiest to automate are repetitive, text-heavy, rules-based, and already digital. Customer support scripts, documentation, records processing, market summaries, and coding assistance all fit that pattern. Anthropic’s separate Economic Index also shows AI adoption moving quickly but unevenly: 40% of employees report using AI at work, up sharply from 2023, while enterprise API use is even more automation-heavy than consumer chatbot use.
Not every occupation is on the same curve.
The least exposed jobs remain those tied to physical environments, in-person service, equipment handling, or tasks that rely on embodied judgment. Cooks, motorcycle mechanics, lifeguards, bartenders, and dishwashers sit near the bottom. Anthropic’s researchers put the boundary plainly: “Many tasks, of course, remain beyond AI’s reach from physical agricultural work like pruning trees and operating farm machinery to legal tasks like representing clients in court.”
The demographic pattern is also striking. Workers in the most exposed professions are more likely to be older, female, more educated, and higher-paid. That undercuts the early assumption that AI risk would fall mainly on low-wage routine work. A growing body of research, including a separate Microsoft study of occupational AI applicability, points in the same direction: knowledge work is highly exposed even when whole jobs are not easily eliminated.
For now, the chart is less a layoff map than a map of workflow redesign. The immediate question is not which jobs disappear overnight, but which roles start changing fastest, which entry paths narrow first, and which tasks become supervised machine work instead of human-first work.
