New AI adoption index pinpoints top sectors: finance, CS, arts

A new open-source economic index aims to measure both AI adoption and capability across occupations using publicly available data. Developed by researchers at arXiv (Somerstep, Guha, Srivastava, Sun), the index leverages user-LLM chat data and O*NET task descriptions to replicate studies typically produced by frontier AI labs. The analysis reveals that occupations in finance, computer science, and the arts show the highest AI adoption rates, offering a transparent benchmark for tracking real-world labor integration.

To assess AI capability, the team built a system that generates benchmark scenarios grounded in O*NET occupations, tasks, and Model Context Protocol (MCP) servers. They tested Kimi-k2.5 using an OpenAI agents SDK harness on scenarios across nine occupations that frequently appear in the index. The results show that AI correctly executes high-level workflows but often makes errors in granular details, such as specific tool calls, highlighting current limitations in precision tasks.

This work provides a much-needed open alternative to proprietary studies from AI labs, enabling independent verification and broader analysis. By combining adoption data with capability testing, the index offers a holistic view of AI's growing role in the workforce. The findings underscore that while AI is rapidly being adopted in knowledge-intensive fields, its reliability in executing detailed occupational tasks remains a work in progress.