On March 27, 2025, Anthropic released the second report from its Anthropic Economic Index, analyzing how Claude.ai was used in the period immediately following the launch of Claude 3.7 Sonnet. The new analysis examines roughly 1 million anonymized Claude.ai Free and Pro conversations collected over the 11 days after the model’s release.
Measurement approach
Anthropic used its privacy-preserving analysis tool Clio to map each conversation to tasks in the U.S. Department of Labor’s O*NET database (about 17,000 tasks). The report inspects patterns across occupations and higher-level occupational categories, and it focuses on interaction modes (e.g., Learning, Task Iteration, Directive) as well as usage of the model’s new "extended thinking" mode. Most of the conversations in the sample came from Claude 3.7 Sonnet, which is the default on Claude.ai and Anthropic’s mobile app.
Key findings
-
Modest increase in coding, education and science: compared with the prior sample, the share of usage in several occupational categories rose modestly, with the computer and mathematical category showing the largest absolute increase (Anthropic reports about a +3% change for that category).
-
Extended thinking is used mostly for technical tasks: the new extended thinking mode is concentrated in technical and technical-creative tasks. Tasks linked to computer and information research scientists show nearly 10% use of extended thinking, software developers about 8%, multimedia artists ~7%, and video game designers ~6%.
-
Augmentation versus automation: overall augmentative uses remain at about 57% of interactions. The share of Learning interactions grew from roughly ~23% in the earlier sample to ~28% in this one.
-
Interaction patterns by task and occupation: Anthropic released breakdowns at task and occupation levels. For example, tasks tied to copywriters and editors show the highest share of Task Iteration (Anthropic cites copywriters around ~58%), while tasks tied to translators and interpreters show high levels of Directive behavior, where the model completes work with minimal human involvement. Librarians lead the Learning category at approximately ~56%.
A 630-category bottom-up taxonomy
To complement O*NET’s top-down taxonomy, Anthropic published a bottom-up dataset built from observed user activity on Claude.ai. The dataset contains 630 granular clusters, each accompanied by descriptions, prevalence metrics, and automation/augmentation breakdowns, organized into three hierarchical levels. Example clusters include assistance with water management systems, creating physics-based interactive simulations, guidance on battery technologies and charging systems, and help with time zone handling in code and databases.
Anthropic positions this bottom-up taxonomy as a way to surface use cases that a top-down O*NET mapping might miss, since general-purpose models are used for tasks not necessarily present in traditional labor-statistics taxonomies.
Data availability and methods notes
All datasets used for these analyses are freely available for download on Anthropic’s Hugging Face page, including the per-task thinking-mode fractions mapping. The report’s appendix contains additional charts and methodological detail. Anthropic notes a few methodological changes from its prior report: it no longer filters conversations for occupational relevance (it only excludes conversations flagged by safety classifiers), and the analysis uses Claude 3.7 Sonnet uniformly instead of the prior Claude 3.5 Sonnet.
Conclusion
Since the launch of Claude 3.7 Sonnet, Anthropic observes modest shifts toward coding, education and scientific uses, continued predominance of augmentative interactions (57%), and increased Learning interactions (~23% to ~28%). The extended thinking mode is used most frequently in technical domains, and the company has published task- and occupation-level breakdowns as well as a 630-category bottom-up taxonomy to support further research. Anthropic intends to continue tracking these metrics and developing new measures as model capabilities and product usage evolve.
Careers
Anthropic invites applicants for roles including Societal Impacts Research Scientist, Research Engineer, and Economist to work on AI labor-market research.
