Anthropic Economic Index: Insights from Claude 3.7 Sonnet

Anthropic has released its second research report via the Anthropic Economic Index, analyzing 1 million anonymized Claude.ai conversations following the launch of Claude 3.7 Sonnet. The findings indicate a modest increase in the share of usage for coding, education, and the sciences, while the overall balance between AI augmentation and automation remains stable.

Shift in Occupational Usage Patterns

Since the introduction of Claude 3.7 Sonnet, there has been a measurable increase in the proportion of usage within specific occupational categories. The most significant absolute increase occurred in computer and mathematical occupations (+3%), with notable percentage increases also seen in education and the sciences.

Anthropic attributes these shifts to three potential factors:

The improved coding benchmarks of Claude 3.7 Sonnet.
The ongoing diffusion of AI across the broader economy.
The application of coding capabilities to non-coding domains or unexpected improvements in general model capabilities.

Usage of Extended Thinking Mode

Claude 3.7 Sonnet's "extended thinking" mode is primarily utilized for technical and creative problem-solving. Analysis of the data shows that tasks associated with specific high-skill roles lead in the adoption of this feature:

Computer and Information Research Scientists: ~10% usage
Software Developers: ~8% usage
Multimedia Artists: ~7% usage
Video Game Designers: ~6% usage

Augmentation vs. Automation Trends

AI usage continues to be predominantly augmentative, comprising 57% of total usage. While the overall ratio has remained constant, the internal composition of these interactions has shifted, specifically with "learning" interactions (where users seek information or explanations) increasing from approximately 23% to 28%.

Variation by Occupation

The balance between augmentation and automation varies significantly depending on the occupational category:

Highly Augmentative: Community and Social Service tasks (including education and guidance counseling) approach 75% augmentation.
Balanced: Production and computer/mathematical occupations skew closer to a 50-50 split between augmentation and automation.

Interaction Mode Examples

Specific occupations exhibit distinct interaction patterns based on the nature of their tasks:

Task Iteration: Copywriters and editors show the highest levels of task iteration, where humans and the model co-write content.
Directive Behavior: Translators and interpreters show some of the highest levels of directive behavior, where the model completes the task with minimal human involvement.
Learning: Librarians show the highest proportion of learning interactions at approximately 56%.

Bottom-Up Usage Taxonomy

To capture use cases that fall outside the U.S. Department of Labor's O*NET database, Anthropic introduced a bottom-up taxonomy consisting of 630 granular clusters. This dataset identifies specific, niche applications of Claude.ai, such as:

Creating physics-based simulations with interactive visualization.
Providing guidance on battery technologies and charging systems.
Troubleshooting font selection and implementation.
Handling time zone logic in code and databases.
Assisting with water management systems and infrastructure projects.

Methodology and Data Availability

Anthropic used its privacy-preserving analysis tool, Clio, to map conversations to O*NET tasks. For this report, Clio was powered by Claude 3.7 Sonnet, which Anthropic notes increased classification accuracy over previous versions. The analysis filtered out conversations flagged by safety classifiers but did not filter based on occupational relevance to preserve data for the bottom-up taxonomy.

All datasets used for these analyses, including the O*NET task-to-thinking-mode mapping and the bottom-up taxonomy, are freely available for download on Hugging Face.