AI adoption has become a paradox: organizations deploy tools like GitHub Copilot and Claude Code, yet struggle to translate individual productivity gains into systemic learning. Ethan Mollick’s research reveals that while engineers automate tasks or accelerate coding, the knowledge doesn’t scale beyond isolated teams. The “messy middle” of adoption—where AI use is widespread but disconnected from organizational processes—leaves leadership scrambling to measure ROI.

A $2 million spend on Anthropic last year yields little more than license usage metrics, prompting calls for better frameworks to capture *actual learning*.