Work communities are the engines of collective intelligence

I keep running into the same mistake, dressed up in different clothes: organizations treat community like a perk. A Slack channel, a monthly town hall, a "culture committee" someone volunteered for in an all-hands. Nice to have, first thing cut when budgets tighten. I think that's exactly backwards, and I think the AI moment is proving it in real time.

Here's the reframe. A community isn't a morale program. It's infrastructure — the actual wiring that intelligence, human or otherwise, moves through inside an organization. Peter Senge spent a career making the case that organizations only learn as fast as the relationships inside them allow information to travel, and Amy Edmondson's research on psychological safety showed why some teams can say the uncomfortable true thing out loud and others can't, and it isn't intelligence that separates them, it's trust. Etienne Wenger's work on communities of practice went further: expertise doesn't really live in individual heads so much as in the ongoing conversation between people doing similar work. Take away the conversation and the expertise doesn't transfer, it just decays in place, one role departure, org change, retirement, or resignation at a time.

I think about this through the same lens as corporate epigenetics — the idea that an organization's structural DNA barely changes, but its environment determines which parts of that DNA actually get expressed. A community is one of the strongest environmental levers available. The org chart says who reports to whom. The community determines who actually talks to whom, who trusts whose judgment enough to act on it without a memo, and how fast a good idea in one corner of the company reaches the corner that needed it. Woolley and Malone's research on collective intelligence backs this up directly: they found that a group's collective intelligence has less to do with the individual IQs in the room and more to do with how evenly people participate and how well they read each other — in other words, the community dynamics, not the credentials.

This is exactly the terrain humanᴬᴵ has to operate in. AI doesn't activate because it's installed. It activates because someone trusted enough to go first went first, and someone else was watching. That's the whole premise behind the trust transfer model—adoption spreads person to person, through relationships people already trust (see example illustration below), not through a rollout email. Adaptation velocity—how fast people actually change how they work—turns out to correlate less with how good the training was and more with how strong the surrounding community already is. Strong community, fast adaptation. Weak or fragmented community, and even excellent tools stall out waiting for someone to go first.

Google's Project Aristotle looked at this from the inside of hundreds of their own teams and landed on a similar answer: the highest-performing teams weren't the ones with the most talented individuals, they were the ones where people felt safe taking a risk in front of each other. That's a community property, not an individual one, and it's not something you can hire your way into—you have to build it deliberately, the same way you'd build any other piece of infrastructure.

Which is where I think HR's real opportunity sits. Not owning AI adoption as much as it is supporting the thing that actually makes AI adoption possible. HR has always been the function closest to the relationships, the trust, company culture, the informal network that McKinsey's organizational network analysis keeps finding is a better predictor of where information actually flows than the org chart is. That's not a new mandate. It's the existing one, pointed at a new problem. If HR shows up to the AI conversation asking "how do we drive license utilization," it's fighting IT's battle with IT's weapons. If it shows up asking "how do we strengthen the communities people actually trust enough to learn from," it's playing to a strength nobody else in the building has.

I'd think about building that strength in three overlapping stages: first, the belonging stage, where people simply start recognizing each other as part of something shared—a team, a guild, a cohort going through the same change at the same time. Second, the practice stage, where that recognition turns into actual exchange: people start showing each other real work, not polished case studies, and the exchange gets specific enough to be useful. Third, the compounding stage, where the community starts generating insight nobody in it had individually — the thing Woolley and Malone were measuring, and the thing that makes the whole exercise worth the investment in the first place.

None of this happens on the timeline of a software rollout, and none of it shows up cleanly on an engagement survey. But it's the actual mechanism. The tools will keep changing. Whether people trust each other enough to learn from each other in public won't, and I think that's the truest lever anyone—HR included—has for making AI activation real instead of theoretical.

References

Senge, P. (1990). The Fifth Discipline: The Art and Practice of the Learning Organization.
Edmondson, A. (1999). Psychological Safety and Learning Behavior in Work Teams.
Wenger, E. (1998). Communities of Practice: Learning, Meaning, and Identity.
Woolley, A., Chabris, C., Pentland, A., Hashmi, N., & Malone, T. (2010). Evidence for a Collective Intelligence Factor in the Performance of Human Groups. Science.
Davenport, T. (2005). Thinking for a Living: How to Get Better Performance and Results from Knowledge Workers.
Castells, M. (1996). The Rise of the Network Society.
Microsoft. (2024–2025). Work Trend Index Reports.
Happe, R., & The Community Roundtable. State of Community Management Research.
Malone, T. (2018). Superminds: The Surprising Power of People and Computers Thinking Together.
Devonport research domains 1–8 (2026).