75% of what computer programmers do can now be handled by AI. That's the headline number from Anthropic's "Labor Market Impacts of AI" study, published March 5, which combines real Claude usage data with occupation databases to map which jobs face the most exposure to AI automation.
But before anyone panics: the study's most important finding is the gap between what AI could do and what it actually does. Computer and math occupations show 94% theoretical AI capability. Actual coverage? Just 33%.
The Most Exposed Jobs
Anthropic created a new metric called "observed exposure" that combines three data sources: the O*NET occupation database (800+ US jobs), actual Claude usage patterns from August to November 2025, and theoretical LLM capability scores. The top 10 most exposed jobs by task coverage:
- Computer programmers - 75%
- Customer service representatives - 70%
- Data entry keyers - 67%
- Medical record specialists - 67%
- Market research analysts - 65%
- Sales representatives - 63%
- Financial analysts - 57%
- Software QA analysts - 52%
- Information security analysts - 49%
- Computer support specialists - 47%
The pattern is clear: jobs that involve moving information around ("bits") rather than physical objects ("atoms") sit squarely in AI's crosshairs.
Who's Safe
About 30% of all workers have zero AI task exposure. These are roles requiring physical presence, manual dexterity, and unpredictable real-world environments: cooks, motorcycle mechanics, lifeguards, bartenders, groundskeepers, maintenance workers. If your job requires you to physically be somewhere and react to messy, real-world situations, AI isn't coming for it anytime soon.
The Demographics Are Uneven
The workers most exposed to AI disruption earn 47% more than average, are 16 percentage points more likely to be female, and are 3.9 times more likely to hold graduate degrees. This isn't a blue-collar automation story. It's a white-collar one, concentrated among educated, higher-earning knowledge workers.
No Employment Cliff Yet
The study found no systematic increase in unemployment among highly exposed workers since ChatGPT launched in late 2022. The one warning sign: job-finding rates for young workers (ages 22-25) in exposed occupations dropped 14%, and hiring into those roles declined about 0.5 percentage points per month. Entry-level knowledge work is where the squeeze is showing up first.
Anthropic's own scenario modeling paints a wide range of outcomes. A worst case where all top-10% exposed workers lost their jobs would spike unemployment from 3% to 43% in that group. A more realistic "Great Recession" scenario doubles it from 3% to 6%.
The researchers note that "AI is far from reaching its theoretical capability: actual coverage remains a fraction of what's feasible." The gap between what AI can theoretically do and what organizations have actually deployed it to do is enormous. That gap is where the timeline lives. The capability exists. The adoption is what's lagging, and that's driven by organizational inertia, trust, regulation, and integration costs rather than technical limitations.