Workers See Value of AI but Companies Not Benefiting
Ethan Mollick states in his One Useful Thing Substack that a large percentages of workers are using AI individually (up to 40% by April 2025), but organizations aren't seeing proportional performance gains.
The disconnect occurs because individual AI use doesn't automatically translate to organizational improvement without deliberate structural changes. Mollick identifies the following three groups, and recommends strategy for each, to help address that.
3-Part Framework
Leadership: Must provide both urgency and vision. Beyond recognizing AI's importance, leaders need to paint a vivid picture of what AI-powered work will actually look like. Mollick’s post emphasizes that workers aren't motivated by performance metrics alone. They need clear images of future work conditions, job security and reward structures. Leaders must also anticipate how AI will replace specific tasks within jobs rather than entire roles.
The Crowd: Represents employees who naturally experiment with AI. The author notes that while official AI adoption often maxes out around 20%, over 40% of workers admit to using AI privately, reporting significant productivity gains. This creates a "Secret Cyborg" problem where valuable innovations remain hidden due to unclear policies, fear of punishment, or concerns about job security.
The Lab: A centralized but agile group focused on building and testing AI solutions. Their responsibilities include:
Rapidly distributing successful AI prompts and solutions from The Crowd
Creating organization-specific AI benchmarks (since standard benchmarks often don't reflect real work tasks)
Building experimental systems that don't work yet but could with future model improvements
Creating "provocations" or demos that help people viscerally understand AI's transformative potential
Key Strategic Insights
Traditional approaches of outsourcing innovation to consultants or enterprise software vendors won't work for AI because "nobody has special information about how to best use AI at your company." Organizations must develop internal capabilities to experiment and learn faster than competitors.
AI fundamentally changes work bottlenecks. When research takes minutes instead of weeks, "the bottleneck isn't the research anymore, it's figuring out what research to do." This requires rethinking organizational structures, processes, and goals built around human-only intelligence.
Significant barriers for company-wide AI adoption including institutional inertia and regulatory concerns. Many companies still ban AI use for "outdated privacy reasons" despite major models not training on enterprise data. This creates more risk through shadow AI use than permitting controlled experimentation.
Summary
Read Mollick’s full post here.