Individual Differences in Generative AI Effectiveness: An Empirical Puzzle for HRD Research

Professionals with similar backgrounds, given identical generative AI tools, achieve sharply different outcomes. This perspectives piece examines what explains that variance, reviewing evidence on cognitive capacity, domain expertise, metacognition, epistemic calibration, personality, and culture, and argues that existing user typologies describe these differences without predicting their consequences. It then extends the puzzle in a direction the static account misses: the same individual differences that shape AI effectiveness are themselves reshaped by sustained AI use, as cognitive offloading, deskilling, and upskilling inhibition operate differently across the expertise distribution. Recent evidence, including deskilling among experienced clinicians and suppressed learning among students using unguarded AI, indicates that assisted performance and independent capability can move in opposite directions. For human resource development, this reframes AI integration from a uniform training problem into one of matching support to user profiles and protecting the conditions under which capability forms. The article closes with a research agenda that prioritizes longitudinal designs for tracking skill development and knowledge retention as AI use accumulates over time.