Anthropic Economic Index: New building blocks for understanding AI use
Published: January 15, 2026
Overview
Anthropic has released its fourth Economic Index report, introducing "economic primitives"—five foundational measurements designed to track AI's economic impacts over time. These primitives measure task complexity, skill level, purpose (work, education, or personal), AI autonomy, and success rates.
Key Findings
Task-Level Performance
The research reveals that Claude accelerates more complex tasks most significantly. Tasks requiring college-level education (16 years of schooling) showed 12x speedup on Claude.ai, compared to 9x for high school-level tasks (12 years). Success rates vary by complexity: Claude completes college-level tasks successfully 66% of the time versus 70% for simpler work.
Time Horizons for Task Completion
Claude demonstrates different capabilities across platforms. On Claude.ai, the model achieves 50% success rates on tasks requiring approximately 19 hours of human effort, whereas API data shows 3.5 hours, and METR benchmarks indicate 2 hours. The methodology differences account for this variance—users can break complex tasks into smaller steps, creating feedback loops.
Geographic Variation in Usage
Claude's application patterns differ significantly by economic development stage. Higher GDP-per-capita countries use Claude more for work and personal purposes, while lower-income nations emphasize educational applications. This reflects an adoption curve pattern common in technology diffusion.
Occupational Coverage
AI coverage has expanded from 36% of jobs (January 2025) to 49% when measuring task exposure. However, adjusting for success rates provides a different picture—some roles like data entry keyers and radiologists show heavier AI impact than raw task coverage suggests.
Skill-Level Distribution
Claude handles higher-skilled tasks disproportionately. Tasks requiring 14.4 years of education (associate degree level) represent a larger share of AI usage than the economy-wide average of 13.2 years. Removing these tasks would create a deskilling effect across professions like technical writing and teaching.
Productivity Impact
The analysis revises earlier productivity estimates. Task speedup alone suggests 1.8 percentage-point annual productivity gains. Accounting for task reliability, estimates decline to 1.2 percentage points for Claude.ai and 1.0 percentage points for API usage. A 1-percentage-point improvement would restore US productivity growth to late-1990s levels.
Interaction Patterns
Augmentation (52% of conversations) now exceeds automation (45%) as the dominant interaction mode on Claude.ai, representing a reversal from August data but continuing a long-term upward trend for automation usage.
Geographic Distribution
AI adoption remains concentrated in the US, India, Japan, the UK, and South Korea. Within the US, distribution has become more equitable, with modeling suggesting nationwide equalization within 2-5 years if current trends persist.
Methodology
The analysis examines 1 million Claude.ai conversations and 1 million API transcripts using privacy-preserving methods. The primitives framework enables tracking whether automation becomes more prevalent as models improve, potentially shifting consumer tasks toward enterprise applications.
Implications
The report establishes a baseline for comparing future surveys. As Claude improves, tasks are expected to increase in complexity, and successful automation may transition consumer use to business applications, signaling downstream economic effects.