Anthropic Just Mapped Which Jobs AI Is Actually Changing

Everyone has an opinion about which jobs AI will eliminate. Most of those opinions are based on vibes. This week, Anthropic — the company behind Claude — published something more useful: actual data on which jobs AI is already changing, measured by what real workers are doing with it right now.

The research, authored by Maxim Massenkoff and Peter McCrory, introduces a metric called “observed exposure.” Instead of asking “could AI theoretically do this task?” it asks “are people actually using AI for this task in their daily work?” The answers are surprisingly different.

The gap between theory and reality

Here’s the headline number that should change how you think about AI and careers: for computer and math workers, AI could theoretically handle 94% of their tasks. In practice? Only 33% are actually being done with AI assistance. That’s a 61-percentage-point gap between what’s possible and what’s happening.

Office and administrative roles show a similar story — 90% theoretical capability, but a fraction of real-world adoption.

This gap is the most important finding in the report. It means we’re still in the early innings. The disruption hasn’t fully hit yet. And the people who learn to close that gap — who actually know how to use AI tools effectively in professional workflows — have a massive advantage right now.

Who’s most exposed

The research ranks occupations by observed exposure. The top 10:

1. Computer programmers (75%)
2. Customer service representatives (70%)
3. Data entry keyers (67%)
4. Medical record specialists (67%)
5. Market research analysts and marketing specialists (65%)
6. Sales representatives (63%)
7. Financial and investment analysts (57%)
8. Software quality assurance analysts (52%)
9. Information security analysts (49%)
10. Computer user support specialists (47%)

Notice a pattern? These aren’t the jobs pundits usually predict AI will eliminate. They’re skilled, white-collar, well-paid positions. The workers in the most exposed occupations earn 47% more than their zero-exposure counterparts and are nearly four times more likely to hold a graduate degree.

Meanwhile, 30% of all workers have zero AI exposure. Cooks, mechanics, bartenders, lifeguards, groundskeepers — jobs that require physical presence. No LLM is pouring drinks or mowing lawns.

The entry-level squeeze is real

The most concerning finding hits young workers hardest. Among 22-to-25-year-olds, the monthly job-finding rate in high-exposure occupations has dropped roughly 14% since ChatGPT launched. A related study found a 16% employment decline in that same age group.

This matters because entry-level roles — the ones that train you, that build your professional network, that turn a degree into a career — are exactly the roles most affected. When Anthropic CEO Dario Amodei warns that AI could “disrupt half of entry-level white-collar work,” he’s not being dramatic. The data backs it up.

If you’re 18 and planning your career, this should reshape your thinking. The traditional path — get a degree, land an entry-level job, work your way up — has a new variable. The entry-level rung of the ladder is getting thinner.

What this actually means for your career

The report’s authors are careful to note that we haven’t seen mass unemployment in exposed occupations — yet. But the BLS employment projections tell a story: every 10-percentage-point increase in AI coverage correlates with 0.6 percentage points lower projected job growth over the next decade.

The takeaway isn’t “avoid AI-exposed fields.” That’s nearly impossible for knowledge workers. The takeaway is: become the person who uses AI, not the person AI replaces.

Workers with AI skills already earn a 56% wage premium. Prompt engineering shows the highest salary bump. But the real power move is combining domain expertise with AI fluency — a financial analyst who understands machine learning, a marketer who can build automated content pipelines, a healthcare worker who can interpret AI-generated diagnostics.

That intersection — deep domain knowledge plus AI capability — is where the jobs are growing, where the salaries are highest, and where the 94%-to-33% gap creates opportunity instead of anxiety.

The question isn’t whether AI can do your job. It’s whether you can do your job with AI better than someone who can’t.

The window is open — but closing

That 61-point gap between theoretical and observed exposure? It won’t last forever. As tools get easier to use, as companies invest in AI training, as a generation of AI-native workers enters the workforce, the gap will close. The advantage goes to people who move now.

This isn’t about becoming a machine learning engineer. It’s about understanding how AI tools work, when to use them, how to evaluate their output, and how to integrate them into professional workflows. Those are learnable skills. And the sooner you learn them, the bigger your head start.