Individual AI productivity gains (Copilot, Claude, Cursor) rarely become organizational learning without explicit feedback paths connecting loops to shared capabilities.
Key Takeaways
Mollick’s Leadership/Lab/Crowd frame: the Lab must convert individual discoveries into shared practices, but most companies lack this translation layer.
The “messy middle” hits when AI adoption is uneven, partially hidden, and not connected to organizational learning – the adoption unit shrinks to the loop inside the work.
Token-to-output metrics (pull requests, prompt counts) miss the point; token-to-learning measures which decisions improved, which root causes got sharper, which patterns became reusable.
Three needed capabilities: Agent Operations (control, audit, permissions), Loop Intelligence (which loops produce learning vs. sprawl), and Agent Capabilities (distributing useful skills without monolithic agents).
A Loop Intelligence Hub instruments intent, agent output, verification, and human intervention to produce actionable signals: enablement backlogs, capability radars, governance gaps.
