Observed exposure quantifies how much AI is actually automating tasks in professional settings, revealing a significant gap between theoretical capability and real-world use. The new framework combines task-level data from O*NET with Anthropic's Economic Index usage metrics, weighting automated tasks more heavily than augmentative ones. Anthropic's Claude currently automates only 33% of tasks in computer and math roles, despite theoretical coverage of 94%, highlighting the technology's current limitations. This measure shows no systematic unemployment increase for exposed workers since late 2022, though younger worker hiring in these roles has slowed.

While AI theoretically could displace many jobs, actual usage remains far behind. The study finds occupations with higher observed exposure are projected to grow less through 2034, with workers in these roles typically older, more educated, and higher-paid. The framework's value lies in identifying vulnerable jobs before displacement becomes visible, though it may not capture all AI's labor market effects like physical tasks or legal representation.

This approach provides a crucial baseline for tracking AI's economic impact, emphasizing that current evidence shows limited disruption despite widespread concern. The findings suggest policymakers should monitor exposure metrics rather than relying on outdated models of job offshorability or robot displacement.