Methodology

Research Foundation

Our scoring model is calibrated against a March 2026 study by Anthropic economists Maxim Massenkoff and Peter McCrory, which introduced a new measure of AI labor market impact called observed exposure. Rather than estimating what AI can theoretically do, the study measured what AI is actually doing in professional settings today — by analyzing real-world usage patterns across hundreds of occupations.

The findings reframe the conventional wisdom about who is most at risk. The highest observed AI exposure falls not on manufacturing workers or low-wage service workers, but on white-collar, knowledge-economy roles. Computer programmers lead with 75% observed task coverage.

Equally counterintuitive: graduate degree holders are nearly four times more likely to be in a highly exposed occupation than workers with no AI exposure (17.4% vs. 4.5%). Advanced education and high compensation are not shields — they correlate with the kinds of structured cognitive work that AI is most capable of augmenting or replacing.

Observed Exposure vs. Theoretical Capability

A critical distinction runs through our scoring model: the gap between what AI can theoretically perform and what is actually being deployed today. Computer and math occupations show 94% theoretical feasibility but only 33% observed usage. Legal occupations show 80% theoretical but just 15% observed.

Tenure Index scores reflect both dimensions. A role with high theoretical capability and high observed usage receives a near-term risk rating. A role with high theoretical capability but low observed usage receives a medium-term rating — the adoption gap creates a buffer, but not immunity.

Occupational Data Sources

Every assessment begins with the Occupational Information Network (O*NET), maintained by the U.S. Department of Labor's Employment and Training Administration. O*NET documents hundreds of standardized descriptors for over 900 occupations, including detailed task lists and work activity classifications.

When a user enters a job title, our system performs a fuzzy match against the O*NET database to identify the closest standardized occupation. The matched occupation's task statements and work activity data are retrieved and passed to our analysis model.

Scoring Model

The 8–96 Range

Tenure Index deliberately constrains scores to a range of 8 to 96. A score of zero would imply complete immunity from AI-related change — an untenable claim. A score of 100 would imply total and imminent displacement — also untenable, as economic, regulatory, social, and institutional factors consistently modulate adoption speed.

This epistemic humility is a feature. Workforce analysis that communicates false precision misleads the people it purports to serve.

Limitations

• O*NET task data reflects median occupational activities; significant variation exists within any occupation.
• AI capabilities are advancing at a pace that makes medium-term predictions inherently uncertain.
• The gap between theoretical AI capability and observed deployment is substantial and variable.
• Displacement timelines are affected by firm-level adoption decisions, regulatory frameworks, and labor market conditions.
• Our analysis does not account for occupational evolution — the ways job roles often adapt rather than disappear.