Who Delegates to AI? Evidence from 53,000 Agent Configurations

Recent research introduces a new metric, 'delegated exposure,' to assess the actual adoption of AI by workers, rather than just the potential for AI to perform tasks. This metric, operationalized as the Agentic Adoption Index (AAI), measures how closely an occupation's tasks align with AI agent routines already built and shared by practitioners. The study leveraged approximately 53,000 AI agent skill specifications and 18,000 O*NET task statements to develop this index.

Why it matters

Understanding the actual adoption of AI through 'delegated exposure' is critical for strategic workforce planning and technological investment decisions. It shifts the focus from theoretical AI applicability to empirical AI integration, providing a more accurate picture of human-AI collaboration and its impact on productivity and organizational structures.

Key insights

• Traditional measures of AI exposure indicate where AI could perform tasks, but not whether workers have actually adopted it.
• The concept of 'delegated exposure' quantifies instances where a worker has integrated AI into their workflow to perform a task.
• The Agentic Adoption Index (AAI) is proposed to measure delegated exposure by comparing occupation tasks to existing AI agent routines.
• The study analyzed 53,000 agent skill specifications from the Manus Skills Marketplace against 18,000 O*NET task statements.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2608.20425