Anthropic Research has published the second update from its Anthropic Economic Index, an initiative designed to track AI's impact on labor markets and the economy. This report focuses on usage data from Claude.ai after the introduction of Claude 3.7 Sonnet, which includes strengths in agentic coding and a new "extended thinking" mode.
The analysis covers 1 million anonymized Claude.ai Free and Pro conversations over 11 days following the model's launch. Claude 3.7 Sonnet was the default model during this period. The report also introduces a new bottom-up taxonomy of Claude.ai usage, comprising 630 granular categories.
Key Points
- The Anthropic Economic Index is a regular initiative to understand AI's effects on labor markets and the economy.
- The second report analyzes 1 million anonymized Claude.ai Free and Pro conversations over 11 days after Claude 3.7 Sonnet's launch.
- Claude 3.7 Sonnet's "extended thinking" mode is primarily used in technical and creative problem-solving contexts.
- Computer and information research scientists show almost 10% usage of extended thinking, followed by software developers at around 8%.
- Digital creative roles like multimedia artists (~7%) and video game designers (~6%) also show substantial usage of extended thinking.
- The balance between augmentation (57%) and automation in AI usage remains largely unchanged from previous findings.
- Learning interactions, where users seek information or explanations, increased from ~23% to ~28%.
- A new bottom-up taxonomy of 630 granular usage categories on Claude.ai has been released.
Context
According to Anthropic Research, the Anthropic Economic Index aims to provide data and research on AI's effects on labor markets. The analysis relies on Clio, a privacy-preserving tool that maps each conversation to one of 17,000 tasks in the U.S. Department of Labor’s ONET database. This approach allows for the examination of overall patterns in occupations and high-level occupational categories associated with these tasks. The initial report from the Index, released previously, also used Clio on approximately 1 million conversations with Claude to organize them by occupational task.
This second report notes modest increases in the proportion of usage in occupational categories such as coding, education, and the sciences. While the increase in coding usage was anticipated due to Claude 3.7 Sonnet's improved performance on coding benchmarks, increases in other categories could indicate broader AI diffusion, novel applications of coding, or unexpected model capability improvements. The new bottom-up taxonomy addresses potential gaps in the ONET dataset by revealing use-cases that might be missed by top-down approaches.
Why It Matters
This research provides builders and researchers with concrete data on how Claude 3.7 Sonnet is being used in real-world scenarios, particularly highlighting the application of its new "extended thinking" mode. The release of detailed datasets, including the new bottom-up taxonomy and task-level automation/augmentation data, offers resources for further analysis into AI's evolving impact on various professional tasks and labor markets.
What To Do
- Download and review the freely available datasets from the Anthropic Economic Index for detailed usage patterns.
- Examine the new bottom-up taxonomy of 630 granular usage categories to identify novel or under-recognized AI applications.
- Investigate the dataset mapping ONET tasks to their associated thinking mode fractions to understand where "extended thinking" is most applied.
- Compare the provided automation and augmentation data at the task and occupation levels to assess the balance of AI use in different domains.