While existing literature measures how occupations are exposed to AI based on theoretical capability, a new research paper published on arXiv proposes a different layer: "delegated exposure." This metric records whether a worker has actually committed a task to AI by building it into a workflow, rather than just identifying where AI could potentially perform.
The Agentic Adoption Index (AAI)
The researchers operationalized this concept as the Agentic Adoption Index (AAI). This index measures the alignment between an occupation's tasks and the agentic routines that practitioners have already constructed and shared. The methodology involved embedding approximately 53,000 agent skill specifications from the Manus Skills Marketplace and computing their semantic similarity to 18,000 O*NET task statements.
Key Findings and Demographic Trends
The study highlights three primary findings:
- Concentrations of delegation differ sharply from the occupations identified as most at risk by pre-AI frameworks.
- The AAI tracks what AI is capable of doing more closely than what workers are currently using it for in practice.
- Adoption peaks below the very top of the wage distribution and at the bachelor's degree level, showing a decline at both the lowest and highest extremes.
The High-Education Shortfall
A significant observation is that technical feasibility alone cannot account for the patterns of adoption. Specifically, there is a notable shortfall among the most highly educated occupations. The researchers suggest this may reflect work that resists advance specification or a level of professional discretion regarding the pace of codification. Distinguishing between these factors—whether the work is inherently un-codifiable or if workers are choosing not to codify it—will require repeated measurements over time.
