Everyone has an opinion on which jobs AI will kill first. Now Anthropic has actual data, and the answer is more nuanced than the doomsday crowd wants to admit.
The company published new research on March 5 introducing a metric called “observed exposure,” which tracks how Claude is being used in real workplaces, not just what it could theoretically do. Turns out those two things are very different, and that difference is where millions of jobs are still sitting comfortably.
For coders, customer service reps, and data entry workers, though, the numbers are harder to ignore.
What the Anthropic Economic Index actually measured
To build this, Anthropic’s team pulled from three sources: the O*NET database covering roughly 800 U.S. occupations, Claude’s own usage logs, and a 2023 academic framework that scores whether an AI can cut a task’s completion time in half.
Every job ends up with a coverage score. High score means AI is already doing a real chunk of that job’s tasks. Zero means it hasn’t shown up in the data at all.
Here’s the part that surprises people. AI could theoretically handle 90% of tasks in office and admin roles. But in practice, observed usage covers only about a third of computer and math jobs, which are already the most penetrated category. The gap between capability and reality is enormous.
These jobs already feel the pressure of AI
Computer programmers top the list at 75% task coverage. Claude is being used heavily for coding, and that usage is leaning toward full automation, not just helping programmers work faster.
Customer service reps come in second. Their core tasks are increasingly showing up in first-party API traffic, which is a technical way of saying companies are quietly replacing human agents with AI pipelines.
Data-entry workers are third at 67% coverage. Reading documents and entering data is exactly what AI does quickly and cheaply, and businesses have noticed.
Other high-exposure occupations include:
- Financial analysts, whose modeling and number-crunching work is heavily covered
- Office administrators, facing 90% theoretical exposure, even if real adoption still lags
- Computer and math workers broadly, where observed exposure sits at 33% and climbing
The Bureau of Labor Statistics backs this up independently. Its employment projections through 2034 show that for every 10 percentage point increase in a job’s AI coverage score, projected growth for that role drops by 0.6 percentage points.
Not catastrophic, but the direction is clear.
The jobs AI still cannot touch
About 30% of U.S. workers score a flat zero. Their tasks simply do not appear in AI usage data at any meaningful level. And before anyone assumes these are low-skill jobs, look at the list.
These are roles built around physical presence, sensory judgment, and reading the room in real time. A language model has no body, no hands, and no instincts. Those still matter.
Zero-exposure occupations highlighted in the research:
- Cooks, whose work involves knife skills, tasting, and plating judgment no model can replicate
- Motorcycle mechanics, who diagnose engines through hands-on inspection
- Lifeguards, whose job is scanning water and executing physical rescues
- Bartenders, who read crowds and social dynamics in real time
- Dishwashers and dressing-room attendants, handling wet, physical, unpredictable tasks
- Agricultural workers pruning trees and operating farm machinery outdoors
- Courtroom lawyers, whose work demands physical presence and live advocacy
The BLS projects steady growth for blue-collar roles through the decade. Health care is adding roughly 40,000 jobs a month, with demand for nurses, therapists, and care workers running well ahead of anything AI is displacing.
KISBENEDEK/AFP via Getty Images
Who faces the greatest AI threat, and what it means for the workforce
Here is where the story gets uncomfortable for a lot of people. The workers most at risk are not who you might expect.
Using Current Population Survey data from just before ChatGPT launched in late 2022, researchers found that the highest-exposure workers tend to be older, more educated, female, and significantly better paid, earning about 47% more than their zero-exposure counterparts.
More Employment:
- Apple CEO Tim Cook drops strong immigration message
- Layoffs in January reach recession-era levels
- Amazon delivers Seattle purge ahead of earnings
Every previous automation wave hit lower-wage workers first. This one is lining up differently, aiming squarely at white-collar professionals who spent years and money building credentials for office-based careers.
That said, there is no unemployment crisis to report yet. The study finds no measurable rise in joblessness among high-exposure workers since ChatGPT launched.
Even a “Great Recession for white-collar workers” scenario, where unemployment in exposed fields doubled from 3% to 6%, would show up clearly in their framework. It has not appeared.
The crack is showing up in hiring instead. Among workers aged 22 to 25, the monthly job-finding rate in high-exposure occupations has fallen roughly 14% since ChatGPT’s arrival.
The drop is barely statistically significant, but it echoes what separate researchers tracking ADP payroll data have been flagging for months: Young people trying to break into exposed fields are finding fewer doors open.
What this means for markets and policy
Investors are already repositioning. Health care and utilities are pulling in overweight allocations from institutional money as white-collar AI fears grow. Software-heavy tech is facing the other side of that trade. The Nasdaq slipped about 1.2% in the session after the report circulated.
Policymakers are moving too, if slowly. Retraining programs targeting trades and care fields are gaining real traction. On the table right now: tax credits for physical-trade apprenticeships, immigration pathways to fill care-sector gaps, and wage subsidies for frontline jobs where AI pressure is low.
Anthropic’s researchers are careful to call this a first step, not a final verdict. They plan to keep updating the coverage measures as usage data evolves and will watch closely whether that youth hiring slowdown deepens.
The honest answer right now is that AI-fueled mass displacement has not arrived. But the early signals are pointing in one direction, and anyone paying attention to where younger workers aren’t getting hired should probably take note.
Related: Wall Street urgently warns software stocks after Anthropic AI move
