It's not factory workers or low-wage service staff who face the greatest AI disruption — it's highly educated, well-paid knowledge workers. And women are overrepresented. The Anthropic Economic Index delivers, for the first time, findings grounded in 4 million real AI conversations — not surveys, not models.
The Counterintuitive Truth About AI and Jobs
Who's really at risk when AI reshapes the labor market? The intuitive answer: low-wage workers, the less educated, people in routine jobs. That answer is wrong — at least when you look at actual usage data.
The Anthropic Economic Index provides, for the first time, an answer grounded not in surveys or theoretical models, but in more than 4 million real conversations with Claude.ai. The finding: the most AI-exposed occupations are well-paid, highly educated — and surprisingly, disproportionately female.
That's not a minor footnote. It inverts the historical pattern of technological disruption entirely.
Klingt interessant?
What Anthropic Actually Measured
The Anthropic Economic Index consists of two research papers. The first — "Which Economic Tasks are Performed with AI? Evidence from Millions of Claude Conversations" (February 2025, Kunal Handa, Alex Tamkin et al., with Dario Amodei and Deep Ganguli) — analyzed usage patterns from Claude.ai conversations. The second — "Labor market impacts of AI: A new measure and early evidence" (March 2026, Maxim Massenkoff & Peter McCrory) — translates those patterns into labor market effects.
The methodological backbone of both papers is Clio, Anthropic's privacy-preserving analysis tool that clusters millions of anonymized conversations into thematic groups without exposing individual content. Those clusters were then matched against the **O*NET database** — a U.S. Department of Labor system covering approximately 20,000 tasks across 800 occupations.
The result of this matching is a new metric: "Observed Exposure" — a composite of real usage data, theoretical feasibility (building on Eloundou et al. 2023), and Bureau of Labor Statistics employment projections. Not who hypothetically could use AI, but who actually does — and for what.
The full dataset is open-source on Hugging Face.
The Most Surprising Numbers
Start with a number that puts everything in context: 37.2% of all Claude queries involve Computer & Mathematical tasks — software development, data analysis, algorithms. By far the largest use category.
But even in this stronghold of AI adoption, Claude covers only 33% of the theoretically possible tasks. The gap between what AI could do and what's actually being used is enormous. We are still in the early stages of adoption.
The second key number: 36% of all occupations use AI for at least 25% of their tasks. Only 4% of occupations use AI for 75% or more. Disruption is already real — but far from universal.
Then there's how AI is being used. The study distinguishes between augmentation (AI works alongside humans) and automation (AI handles the task independently). Result: 57% augmentation vs. 43% automation. In more than half of cases, AI isn't replacing people — it's making them more productive.
That figure matters more than it initially appears. It challenges the prevailing narrative that AI is primarily a replacement engine.
The Top 10 Most Exposed Occupations
The ranking of most AI-exposed occupations is a surprise for anyone who expected factory workers and manual laborers to top the list:
| Rank | Occupation | AI Exposure |
|---|---|---|
| 1 | Computer Programmers | 75% |
| 2 | Customer Service Representatives | 70% |
| 3 | Data Entry Keyers | 67% |
| 4 | Medical Record Specialists | 67% |
| 5 | Market Research Analysts & 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% |
75% for Computer Programmers is a striking number. Boris Cherny, the creator of Claude Code, has said publicly that he expects the "title of software engineer" to "go away" in 2026. That sounds provocative. But if three-quarters of a software developer's tasks are AI-exposed, the provocation begins to make sense.
Second place is equally notable: Customer Service Representatives. 70% exposure, one of the largest employment categories globally. The automation potential here is real and large.
At the other end of the spectrum: Cooks, Mechanics, Lifeguards, Bartenders, Dishwashers. Physical labor remains largely untouched. As Massenkoff & McCrory put it: "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 Paradox
This is where the data becomes genuinely surprising. When you compare the most AI-exposed workers to those with zero exposure, a clear demographic profile emerges:
- +16 percentage points more likely to be female
- +11 percentage points more likely to be white
- Nearly twice as likely to be Asian
- 47% higher earnings on average
- Nearly 4x more likely to hold a graduate degree (17.4% vs. 4.5%)
This contradicts the historical pattern of industrial automation completely. The Industrial Revolution hit low-skill, physical workers, low-wage earners. The AI disruption hits academics, high earners — and, unexpectedly, women disproportionately.
Why women? Because the most exposed occupations — marketing, customer service, documentation, HR, financial analysis — are historically more female-dominated than mechanical engineering or construction. The feminization of white-collar work intersects directly with AI's disruption pattern.
This has policy implications that extend far beyond tech companies. If the IMF is right — and in January 2024 it wrote that AI will "likely worsen overall inequality" — then this time it's not just workers at the bottom of the income ladder who are affected.
Dario Amodei, CEO of Anthropic, was direct about it: AI could replace "up to half of all entry-level white-collar jobs in the next one to five years." Entry-level jobs. The career ladder that universities implicitly promise graduates.
What Has Actually Happened So Far
Enough theory. What can we actually measure as of early 2026?
The Yale Budget Lab (October 2025) reached a clear conclusion: no measurable employment displacement is detectable yet. Overall employment numbers remain stable.
But early warning signals are emerging. A Stanford study (November 2025) found a 16% relative decline in new hires of young graduates in AI-exposed roles. Companies aren't necessarily hiring fewer people overall — but they're hiring fewer junior entrants into the most exposed positions.
In concrete numbers: approximately 55,000 AI-related layoffs were counted in the US in 2025. Amazon: 15,000 positions. Salesforce: 4,000. Accenture, Lufthansa, and others followed. That sounds large — but measured against total US employment of over 160 million, it's still a fraction.
For every 10 percentage-point increase in AI exposure, BLS growth projections fall by 0.6 percentage points. A statistically significant effect — but not a labor market collapse.
Still, the mood is shifting. A Mercer survey of 12,000 employees shows that the share of workers who fear AI-related job loss has risen from 28% (2024) to 40% (2026). Anxiety is outpacing reality — but perhaps that's not entirely irrational. IMF Managing Director Kristalina Georgieva said it plainly at Davos 2026: AI is hitting the labor market "like a tsunami."
What Other Studies Say
The Anthropic Economic Index doesn't stand alone. Several major institutions have published their own analyses in recent years — and the baseline is consistent, if nuanced.
IMF (Davos 2024): 40% of all jobs globally affected; 60% in advanced economies. Some will benefit from AI as a complementary technology; others will be displaced. Net effect: likely more inequality.
Goldman Sachs (2023): Equivalent of 300 million full-time positions affected; 25% of all tasks automatable. But simultaneously: +7% GDP possible — if productivity gains are captured effectively.
World Economic Forum, Future of Jobs 2025: 170 million new jobs created by 2030, 92 million lost — net: +78 million jobs. 86% of businesses expect AI to fundamentally transform their operations.
The synthesis: the disruption is real. But "job killer" is too simple. Whether the net trajectory is positive depends heavily on how companies, policymakers, and individuals respond.
What Organizations and Individuals Should Do Now
Analysis is useful. Drawing the right conclusions is better. Five concrete steps:
1. Know your own exposure
The O*NET database is publicly accessible. Enter your job title and see which of your tasks are theoretically AI-exposed. This creates clarity instead of diffuse anxiety.
2. Develop an augmentation mindset
57% of AI usage is collaborative, not replacing. Those who view AI as a productivity lever rather than a threat are better positioned. The question isn't "Will AI replace me?" but "How do I become more productive with AI?"
3. Build AI fluency now
97% of investors say they negatively evaluate firms that don't upskill their workers on AI. AI literacy is no longer optional — it's a signal of strategic capability and institutional health.
4. Protect new talent pipelines
The Stanford study suggests entry-level hiring is already softening. The long-term consequence: if no junior positions exist, there will be no senior talent in ten years. Cutting new graduate hiring to save costs is optimizing for the wrong variable.
5. Focus on genuinely human value
Client relationships, complex judgment, physical execution, emotional intelligence — these are areas where AI remains structurally weak. Orienting your role in these directions builds a durable advantage that won't be competed away in the next model release.
Conclusion: The Most Important Finding Gets Overlooked
The headlines will focus on the table of exposed occupations. The demographic paradox will spark debate. But the most important number is the most understated one: 57% augmentation.
AI is — at least as of today — primarily not a replacement engine. It's a leverage tool. More than half of actual AI usage serves to amplify human work, not supplant it.
That will change. The direction is clear. But the window is still open: organizations and individuals who understand now how to use AI as a lever — rather than defensively waiting to be disrupted — will be structurally better positioned.
At Nopex, we work with teams every day to operationalize exactly this distinction: AI not as an automation hammer, but as an augmentation tool. The Anthropic study confirms that this is the right framing. And the data comes not from surveys — it comes from 4 million real conversations.
