We introduce the notion of delegated exposure and operationalize it with the Agentic Adoption Index (AAI), which records whether a worker has actually assigned a task to AI within a workflow. Using roughly 53,000 agent skill specifications from the Manus Skills Marketplace, we compute their semantic similarity to about 18,000 O*NET task statements and aggregate the scores at the occupation level to obtain the AAI.
Three key findings emerge. First, occupations with high AAI differ sharply from those traditionally flagged as “high‑risk” in pre‑AI exposure studies. Second, the AAI aligns more closely with what AI is capable of rather than what workers are currently using it for. Third, the AAI peaks at the middle of the wage distribution and at the bachelor‑degree education level, falling off at both the low‑ and high‑ends. While technical availability explains most of the variation, the shortfall among the most educated occupations indicates that feasibility alone cannot account for adoption; the gap may reflect tasks that resist fine‑grained specification or professional discretion over codification speed.
Repeated measurement will be needed to disentangle these factors and track how they diverge over time.
Blogger's Review: This work offers a fresh, workflow‑centric lens on AI adoption, urging policymakers to look beyond mere task replaceability and consider the actual pathways through which AI is embedded in work.