Ethan Mollick identifies a major paradox in enterprise AI adoption: while individual workers report significant productivity gains (some claiming AI "tripled their productivity"), organizations see minimal overall performance improvements.

Four central observations

First, performance gains are real and documented. Studies in Denmark and the United States show workers achieving substantial time savings across varied domains: product development, sales, consulting, and technical roles.

Second, adoption is widespread. Between 30 and 40% of US workers use AI professionally, though official enterprise systems show lower engagement rates (~20%), revealing significant "shadow" usage.

Third, untapped potential exists. Deep research tools, AI agents, and content generation systems can accomplish transformational work well beyond current organizational implementations.

Fourth, organizations are not capturing this value. Individual productivity gains do not automatically translate into organizational improvements without systemic innovation.

The Framework: Leadership, Lab, and Crowd

Mollick proposes a three-component model to resolve this paradox.

Leadership must establish a clear vision of AI's future impact on work and create incentives that encourage transparent adoption rather than hidden usage driven by fear of layoffs or diminished recognition.

The Lab functions as an ambidextrous innovation unit, turning workflows discovered by the crowd into scalable solutions, establishing AI capability benchmarks on the organization's real tasks, and building experimental prototypes that test emerging possibilities.

The Crowd represents experienced workers who organically discover effective AI applications through trial and error, then share successful workflows across the company.

Practical recommendations

Mollick stresses the need to move beyond vague ethical guidelines to establish explicit experimentation zones. Training should be reframed as hands-on experience rather than instruction in prompting techniques. Organizations must build benchmarks measuring AI performance on their real tasks.

He also highlights the importance of addressing incentive structures that discourage workers from revealing AI-assisted productivity, and of fundamentally rethinking the purpose of tasks when efficiency gains eliminate previous bottlenecks.

Strategic vision

Mollick's critical insight: competitive advantage belongs to organizations that learn fastest, not those that wait for perfect clarity. "The time to start is not when everything becomes clear—it's now, while everything is still messy and uncertain."