Hadar BismutNot long ago I read a fascinating study published in 2024 (opens in a new tab) on how organizations successfully adopt AI. The findings challenge many of our basic assumptions about how technological knowledge spreads in organizations, and explain why most AI initiatives fail.
What doesn't work (and why we all do it)
The myth of natural diffusion
Many people believe that if one team runs a successful AI pilot, the knowledge will naturally "leak" to other teams. The study shows that this simply doesn't happen. Knowledge doesn't spread randomly through hallway conversations. It requires intentional, structured channels of distribution.
The study identifies this phenomenon as "closed networks": expert communities that keep knowledge inside and don't pass it on. They don't do it on purpose, of course. It happens mainly because there's no infrastructure for sharing.
The "top-down mandate" failure
The second belief is that if leadership announces an AI policy or orders a broad "rollout" of the technology, the organization will move forward. The study shows that top-down decisions don't guarantee internalization or action. Often it stays at the slide-deck level.
Why? Because between the leadership announcement and implementation in the field, there's a huge gap in understanding, skills and concrete business reasons. An employee on the ground doesn't know how to turn generic ChatGPT into a tool that helps in their daily work.
What does work: the power of cognitive proximity
The study identifies one strong mechanism that does ensure knowledge spreads: peers in cognitive proximity.
That means real connections between peers who share a professional language (terminology, methodologies, tools), a similar technological background, trust, and the opportunity to share practical knowledge. That's where knowledge flows, turns into practice, and from there can spread further.
When people with high cognitive proximity connect, something special happens: the knowledge gets passed along and also adapted to a specific context (here too, Context is king). Instead of generic instructions on "how to use AI," the employee gets practical advice on "how to use AI to solve exactly the problem you have."
Professional guilds are the best human API
This is where guilds come in: internal professional communities that bring people together around a technical expertise or professional field. Guilds are the organizational structure that naturally creates the cognitive proximity this requires.
When a guild member learns how to use AI in the context of their project, they bring that knowledge to the guild. Other members translate it to their own needs, enrich it, and bring more insights back to the community.
It's an evolving learning network, far more than a one-way transfer of information, and it creates a rich knowledge base that's relevant to every participant.
So while the study itself wasn't about guilds or professional communities, its findings echo what I see in the field: if you really want AI knowledge to move beyond the slide deck or a side pilot, let your guilds do the work. They're the best human API an organization has.