The Open Source Economic Index of AI Adoption and Capability

A New Yardstick for the AI Economy

A new preprint on arXiv (2606.26118v1) proposes an open-source economic index designed to measure both AI adoption and AI capability across occupations. The researchers leverage publicly available user-LLM chat data alongside O*NET, the U.S. Department of Labor’s comprehensive database of occupational requirements. By correlating real-world usage patterns with granular task-level descriptions, they aim to quantify not just which jobs are being augmented by AI, but how effectively AI can perform the discrete labor tasks within those roles.

Why This Matters

This work addresses a persistent blind spot in AI economics. Most existing metrics track either raw model performance on benchmarks (e.g., MMLU, HumanEval) or coarse adoption surveys that ask “Have you used ChatGPT?” Neither approach captures the nuanced reality of how AI integrates into actual workflows. The index’s use of O*NET is particularly significant—it breaks occupations into specific tasks, allowing the researchers to measure capability at the task level rather than the job level. A radiologist, for example, might have high AI capability for image interpretation but low capability for patient communication.

The open-source nature of the index is equally important. Proprietary indices from consulting firms or tech companies are often black boxes, making it difficult for policymakers, economists, and AI practitioners to verify or reproduce findings. By grounding the index in publicly available chat data and standardized government datasets, this work enables transparent, replicable analysis of how AI is reshaping labor markets in real time.

Implications for AI Practitioners

For developers and product managers, this index offers a practical framework for prioritizing feature development. If the data shows that AI capability is high for a specific task (e.g., code debugging) but adoption remains low, that signals a usability or trust gap worth addressing. Conversely, high adoption with low capability suggests users are pushing models beyond their current limits—a clear warning to invest in robustness or risk user disappointment.

For enterprise AI strategists, the index provides a more defensible basis for automation decisions. Instead of relying on vendor claims or hype, organizations can cross-reference their own workflows against the task-level capability data to identify where AI deployment is likely to yield genuine productivity gains versus where it remains premature.

The methodology also highlights a growing tension: as AI capability improves, the definition of “discrete labor tasks” becomes fluid. What counts as a single task today (e.g., “write a marketing email”) may soon be decomposed into subtasks (drafting, personalizing, A/B testing). Practitioners should watch how the index evolves to handle this granularity, as it will directly impact which roles are considered automatable.

Key Takeaways

• Transparent measurement matters: An open-source index using public data (O*NET + chat logs) allows for reproducible, verifiable tracking of AI’s economic impact, unlike proprietary alternatives.
• Task-level analysis is superior: Measuring AI capability at the task level rather than the occupation level reveals a more nuanced picture of which work activities are genuinely automatable.
• Actionable for product and strategy: The gap between adoption and capability provides a clear signal for where to invest in UX improvements (low adoption, high capability) or model robustness (high adoption, low capability).
• Granularity will be a moving target: As AI models improve, the definition of a “discrete task” will shift, requiring the index to continuously adapt its decomposition of labor activities.