What has changed since our last report
The latest report is based on data from February 2026, comparing it to previous data from November 2025. The way people use Claude on Claude.ai has become more diverse, with the top 10 tasks making up a smaller share of overall usage. Coding tasks are still common but are shifting from Claude.ai to the API, where they're being broken down into smaller tasks. The average economic value of tasks done on Claude.ai has decreased slightly due to more personal queries and less coding. Globally, usage remains concentrated in a few countries, but within the US, it's becoming more evenly distributed across states. The report is available in PDF format [here](https://www.anthropic.com/research/economic-index-march-2026-report.pdf).
Learning curves
The report explores how users develop habits and strategies over time, particularly how they choose between different AI models like Opus, Sonnet, and Haiku. Users tend to pick Opus for more complex, higher-value tasks. For example, coding tasks are more likely to use Opus than educational tasks. More experienced users, defined as those who've been using Claude for at least 6 months, are better at getting successful responses from Claude. They use Claude more for work, tackle more challenging tasks, and collaborate with Claude more effectively. This suggests that experience plays a significant role in learning to use AI effectively.
Diversification of use cases in Claude.ai
The tasks people ask Claude to perform on Claude.ai have become less concentrated. While coding remains the most common task, associated with Computer and Mathematical occupations, its share has decreased as it migrates to the API. Other tasks like personal queries have increased. The average conversation on Claude.ai now requires slightly less education and has a lower estimated economic value. The way users interact with Claude has also shifted, with a slight increase in 'augmentation' - where Claude complements the user's abilities. This change is partly due to new users signing up and bringing different use cases. The report draws on a sample of 1 million conversations from Claude.ai and the API, using a privacy-preserving system to analyze behavior without revealing individual content.
Emergent automation patterns
Learning to use AI
The report examines how users learn to work with AI over time. More experienced users are not only more successful in their interactions with Claude but also use it more collaboratively and for more complex tasks. They tend to use Opus, the more powerful model, for tasks associated with higher-paid jobs. The data suggests that users get better at using Claude as they gain experience, supporting the idea of 'learning-by-doing'. However, it's also possible that early adopters were more technically inclined, so the results could be influenced by cohort effects or survivorship bias.
The Anthropic Economic Index provides valuable insights into how Claude is being used across the economy. As AI adoption grows, understanding these patterns is crucial for researchers and policymakers. The report shows that while Claude is used for high-value work, its user base is diversifying, and more casual users are emerging. Experienced users are getting more out of Claude, using it more effectively for complex tasks. These findings have implications for how AI might impact the labor market and economy in the future.