The Messy Middle of AI Adoption: When Tools Outpace Organizational Learning

Ethan Mollick's book 'Making AI Work: Leadership, Lab, and Crowd' argues that individual AI productivity gains do not automatically become organizational gains. As companies provision tools like GitHub Copilot, ChatGPT Enterprise, Claude, Gemini, and Cursor, AI use becomes widespread, uneven, and often hidden. This 'messy middle' phase sees teams using AI in vastly different ways—from basic autocomplete to advanced agentic loops—while management struggles to measure ROI beyond token counts. Mollick's framework of Leadership, Lab, and Crowd highlights the challenge: how does learning travel across the organization? Traditional change machinery (communities of practice, brown-bag sessions) is too slow for the pace of AI work. The author introduces 'Loop Intelligence' as a missing feedback path, comprising Agent Operations (control), Loop Intelligence (learning), and Agent Capabilities (distribution). A 'Loop Intelligence Hub' would instrument real work loops to capture learning, but must avoid becoming employee surveillance. The key question shifts from token-to-output to token-to-learning. The next competitive advantage lies in learning velocity—how quickly organizations find patterns, share discoveries, and build backpressure into agentic loops. No definitive playbook exists yet; understanding requires instrumenting work and iterating in the open.