Abstract

This report examines how Claude AI is reshaping the global economy by introducing five foundational "economic primitives", simple measures capturing task complexity, human and AI skill levels, use case categories (work, education, personal), AI autonomy, and task success rates. Analyzing 1 million anonymized conversations from Claude.ai and enterprise API traffic (November 2025), the study reveals striking geographic variation in AI adoption and usage patterns.

Key findings

Key findings indicate that Claude remains concentrated on coding-related tasks (34% of Claude.ai usage, 46% of API traffic), though usage is becoming more evenly distributed across locations, with potential parity achievable in 2-5 years. Takeaways: Prompting is still critical, AI can handle massive tasks if you break them Up, and professional deskilling (here, the loss of professional value, correlated with the loss of skills due to technological changes) is a bigger issue than unemployment.

Globally, GDP per capita strongly predicts AI adoption (0.7% increase per 1% GDP increase), while coursework use dominates lower-income countries and personal use is higher in wealthier nations. AI demonstrates greater productivity gains on higher-education tasks (12x speedup for college-level work vs. 9x for high school level), but success rates decline with task complexity (66% for college-level vs. 70% for basic tasks).

Notably, removing tasks Claude can perform would produce a net deskilling effect across most occupations, as AI tends to automate the more skill-intensive components of jobs. Adjusted for task reliability, labor productivity growth implications are revised downward from 1.8 to 1.0-1.2 percentage points annually over the next decade, with significant variation depending on task complementarity assumptions.

The study concludes that AI's economic impact will be geographically uneven and mediated by existing institutional structures, requiring development of human capital to ensure equitable global benefits. 

In short: Without massive employee training, the economic benefits of AI will remain concentrated in a few countries and companies with the necessary resources and skills.

My take

The study reveals a striking divide: the disparity in AI usage (Claude in this case) between low-GDP countries, which primarily use Claude for training ("coursework"), and wealthier countries, where its use is more personal and professional. However, this divide isn't limited to a clear separation between regions and geographical areas; it's also felt within shared territories. Within the same country, companies (particularly those in the Financial Services sector) don't have the same capacity to adapt to AI adoption, due to disparities in financial and human resources.

The observation of an already existing professional deskilling will only exacerbate the loss of technical skills and, consequently, widen the inequalities between individuals, companies, and countries regarding AI use. Thus, for mid-market players in the financial sector, I believe the risk lies not in AI itself, but in remaining within a "training" framework. Individualized, or even poorly framed, "discovery" (with all the risks that entails*), while large financial institutions shift into "production" mode, this exacerbates a competitive gap that will be difficult to close.

In short: Without structured and consistent investment in employee training, the economic benefits of AI will remain concentrated in a few countries and large institutions with the resources and expertise to train their teams on a massive scale, widening the competitive gap with mid-market financial services players.

The interactive dashboard below allows you to explore in detail the geographical variations in adoption and sectoral disparities.

Report and Dataset: Anthropic Economic Index report: economic primitives, Jan. 15, 2026

*See my opinion piece: Managing AI-related risks in financial services, Jan. 23 2026