Anthropic has released The Anthropic Economic Index, which provides insights into how users are leveraging its online chat tool.

The WEF, WIRED, NYT, and a host of AI media coverage have published observations and opinions from pundits on the potential economic impact of AI tools.

In short, it seems that globalization is still the primary driver of economic changes, more so than the adoption of AI tools in the workplace. But don’t sleep on LLMs; understanding how the market is using these tools and for what tasks can prepare you for the future.

Key Insights

Knowledge Work Dominates AI Usage

Using Clio, this study classified interactions with Claude into broad categories, with Critical Thinking, Writing, Systems Analysis, and Complex Problem-Solving in the top percentage of total interactions. I have personally found brainstorming (complex problem-solving?) to be the most powerful feature of LLMs—it’s not so much that it can DO the task as it can help think through problems and explore topics from various angles.

Chart showing top ten economic tasks performed with Claude according to Anthropic Economic Index

Automation vs. Augmentation Workflows

This paper does a great job attempting to distinguish between AI-automation and AI-augmentation. It also provides a chart showing that very few occupations (~4 percent) have 75 percent or more of daily tasks that can be automated. The death of the radiologist has not happened; the truth is there is much more to any profession than the handful of automatable tasks an LLM can assist with.

automative behaviors vs. augmentative behaviors
Table from whitepaper showing Automative Behaviors and Augmentative Behaviors. Source: Economic Tasks AI Paper. Anthropic.

Even in the case of high automation, LLMs are a string-based nondeterministic statistical model and should not be relied on to complete tasks just because they can.

Ideation and critical thinking are complex multistep processes that LLMs are poorly equipped to manage even with “reasoning” prompt stacking. Further, this whitepaper acknowledges that the analysis is limited to only http://claude.ai/ chat interactions. It excluded API and Business Users from the dataset. So take with a grain of salt Anthropic’s attempts to classify workflows as either fully automated or augmentations of another process.

Citation

Handa, Kumal, Alex Tamkin, Miles McCain, et al. 2025. “Which Economic Tasks Are Performed with AI? Evidence from Millions of Claude Conversations.” White Paper. Anthropic. https://assets.anthropic.com/m/2e23255f1e84ca97/original/Economic_Tasks_AI_Paper.pdf

Significant Revisions

  • Feb 25th, 2025 Originally published on https://www.jsrowe.com with uid 2D86F16C-F228-47C2-8AD2-7D6CA6A1CD94
  • Feb 15th, 2025 Draft created