concept · concepts/Leadership, Lab, and Crowd.md
Leadership, Lab, and Crowd
Ethan Mollick's framework for turning AI's individual productivity gains into organizational ones — by coordinating three roles. The premise: there is no playbook, so advantage goes to whoever learns fastest.
Why a framework is needed
Individual gains don't become organizational gains on their own (Individual vs Organizational AI Gains); capturing them requires organizational innovation, and that capability has "atrophied" from decades of outsourcing to consultants and software vendors. Nobody — not even the AI labs — has special knowledge of how to deploy AI in your context. So the work is internal, fast, and iterative.
The three roles
Leadership — provide urgency and a vivid, concrete picture of the AI-enabled future (Mollick cites colleague Andrew Carton's research: workers change for vivid images of the future, not statements about performance or bottom lines). Decide and signal whether gains become layoffs or growth; anticipate task-level change ("AI replaces specific tasks within jobs, not most jobs"); rethink why tasks are done once their cost collapses. Leadership also owns the fix for Secret Cyborgs.
The Crowd — the employees who discover uses through trial and error. Experienced workers can judge when AI is useful in ways outsiders and AI-savvy juniors cannot, then share workflows. But adoption stalls and gains hide unless Leadership removes the disincentives.
The Lab — a centralized, ambidextrous team (explore the future + exploit the present), staffed from the Crowd, focused on building, not strategy decks. Its outputs:
- Take Crowd prompts/solutions and distribute them fast; build "fast and dirty" products, iterate, release, measure.
- Build your own organizational benchmarks — official benchmarks (trivia/math/coding) don't measure your tasks; "vibe benchmarks" are acceptable (Mollick vibe-benchmarked the Manus agent on a fictional-startup analysis).
- Build stuff that doesn't work… yet — prototype full AI-agent business processes, then plug in each new model to catch the moment it crosses a threshold.
- Build provocations — visceral demos that jolt people past inertia.
The point
Treat AI adoption as an organizational learning challenge, not a technical one — build feedback loops between the three roles so the org learns faster than competitors, and start "now, while everything is still messy."
Connections
- Individual vs Organizational AI Gains — the gap this framework exists to close.
- Secret Cyborgs — the Crowd-level dynamic Leadership must fix.
- The Bottleneck Shift — the deeper reason work itself must be rethought.
- Every — a real org that did this organizational innovation (and captured the gains).
- Personal vs Team Agents — the Lab building shared tools parallels Every's "unit agent" move.