AI

Anthropic Releases Fourth Economic Index Report Introducing New Metrics for Measuring AI’s Workplace Impact

Anthropic has published its fourth Economic Index report, introducing a new framework of “economic primitives” designed to provide deeper insights into how artificial intelligence is transforming work and occupations. The report, which analyzes data from November 2025 using Claude Sonnet 4.5, marks a significant expansion of the company’s ongoing effort to measure real-world AI adoption and productivity impacts.

The new framework introduces five foundational measurements: task complexity, skill level, purpose (work, education, or personal use), AI autonomy, and success rate. These primitives were derived by asking Claude to analyze conversations from both Claude.ai consumer usage and the company’s first-party API, which primarily captures business applications.

Key Findings on Task Performance

The research reveals that AI accelerates more complex tasks to a greater degree than simpler ones. Tasks requiring a high school-level education (12 years of schooling) showed a speedup factor of nine, while those requiring a college degree (16 years) were accelerated by a factor of 12. This pattern suggests that current AI productivity gains are concentrating in occupations requiring relatively high human capital, consistent with evidence that white-collar professionals are more likely to incorporate AI into their workflows.

However, success rates decline as task complexity increases. Claude successfully completes tasks requiring a college degree 66% of the time, compared to 70% for tasks requiring less than a high school education. While this disparity reduces the overall impact, it does not eliminate the relationship between complexity and speedup gains.

The report also examines AI’s effective time horizons—the maximum duration of tasks that AI can successfully complete. According to METR’s benchmark, Claude Sonnet 4.5 achieves 50% success rates on tasks requiring approximately two hours to complete. Anthropic’s own API data shows slightly different results, with Claude achieving 50% success on tasks taking around 3.5 hours, while Claude.ai users see success rates at approximately 19 hours. The discrepancy may be attributed to methodological differences, including users’ ability to break complex tasks into smaller steps on the consumer platform.

Global Usage Patterns

The analysis reveals significant variations in how Claude is employed across different economic contexts. In regions with higher GDP per capita, AI is used more frequently for work and personal applications. Conversely, areas with lower GDP per capita show a greater share of usage for educational coursework, aligning with a broader “adoption curve” pattern where AI use diversifies from educational to personal purposes as economic conditions improve.

These findings complement recent research from Microsoft linking educational AI use to lower per-capita income and leisure-oriented AI use to higher incomes. Anthropic noted its partnership with the Rwandan government and ALX, a technology training provider, which is specifically designed to address this pattern by helping participants transition from educational applications to broader workplace use.

Occupational Impact Analysis

The report shows meaningful growth in AI adoption across occupations. In January 2025, 36% of jobs in the sample had AI used for at least a quarter of their tasks. By the latest analysis period, this figure had risen to 49%.

However, when accounting for success rates—weighted by how frequently workers perform tasks and their duration—the occupational impact picture changes considerably. Some roles, such as data entry positions and radiologists, appear substantially more affected by AI integration than task coverage alone would suggest. Meanwhile, occupations like teaching and software development show relatively less impact when success-adjusted metrics are applied.

The full methodology, including validation details for the economic primitives, is available in chapter two of the complete report.


Source: https://www.anthropic.com/research/economic-index-primitives