Why humans work best in small groups
The barrier isn't AI. It's finding the right four people to think with.
Why humans work best in small groups
The most profitable firms on Wall Street are organised into pods.
At Millennium, Citadel, Point72, Balyasny — the multi-strategy hedge funds that have produced some of the most consistent risk-adjusted returns in finance — capital is allocated not to large teams but to small ones. A portfolio manager and a handful of analysts. Two to six people per pod. Each pod runs its own book, makes its own decisions, lives or dies by its own P&L. Hit your drawdown limit and the pod is cut. Produce, and you get more capital.
The model has been so successful that the industry has restructured around it. Single-manager funds have struggled to compete on returns. Capital has flooded in. Millennium alone runs hundreds of investment teams and over $70 billion in AUM.
I spent ten years on a Delta One desk at Credit Suisse, dealing with these firms daily. The striking thing when you work alongside pods is not the structure on paper but what it produces: a particular density of conversation. Everyone in the pod knows what everyone else is doing. Decisions happen in minutes, not days. There is no “loop me in.” There is no project manager. The pod is the meeting, and the meeting is the pod.
This is not unique to finance, and it is not an accident.
Richard Hackman, the Harvard psychologist who spent forty years studying teams, found that as size grows beyond a single-digit count, coordination costs overwhelm performance gains. The relationships to manage grow quadratically — five people produce ten pairwise connections; ten produce forty-five. Cognitive load scales faster than human capacity to manage it.
The military arrived at the same number from a different direction. US Army fire teams are four people. Special forces operate in four to six person elements. The number isn’t arbitrary; it reflects what humans can coordinate under stress without losing situational awareness. In tech, Jeff Bezos institutionalised the principle as the two-pizza rule: no team so large it cannot be fed by two pizzas. Most two-pizza teams settle at five or six.
The pattern reappears because it reflects two underlying constraints, both about minds rather than work.
The first is what social psychologists call production blocking. In a group of three or four, you speak when you have something to say. By six, you wait your turn. By ten, you mentally rehearse your point while others are talking — which means you have stopped listening. Robin Dunbar’s research on conversation found that gatherings naturally fragment into smaller subgroups once they exceed about four people, because the cognitive overhead of tracking everyone exceeds capacity. Watch any party. The dinner of twelve you imagined is actually three separate dinners of four happening at different ends of the table.
The second constraint is trust. In a group of five, you know how each person thinks. You have watched them be wrong about something and adjust. You know who is overconfident on Mondays and who is worth listening to when they go quiet. By ten, you stop tracking individuals and start tracking factions. The granularity that makes honest exchange possible decays into politeness.
Both constraints scale badly. Communication overhead grows quadratically; human capacity stays flat. The productivity miracle of small groups isn’t really about the individuals. It’s about the bandwidth.
What hedge fund pods discovered, what military fire teams discovered, what two-pizza teams discovered, what every productive small group in history has discovered, is that the format of human collaboration matters more than the talent inside it. You can take six brilliant people and ruin their output by putting them in a room of twenty. You can take six average people and produce extraordinary work by giving them a clear remit, a P&L, and a shared screen.
Most of the conversation about AI has fixated on the wrong question. The question is not what tasks AI will do for us. The more interesting question is what humans uniquely contribute when machines handle the rest. And the answer is becoming visible: the things that depend on shared context, watched judgment, and the kind of rapid, trust-laden exchange that only happens in rooms of five.
The barrier to human progress in this era is not AI. AI is a tool, increasingly cheap and increasingly capable. The barrier is finding the right small group to think with — across the noise of LinkedIn, the dilution of Discord, the broadcast of Twitter, the formality of conferences. Finding four other people who will show up, push back, change their minds, and let you change yours.
That is where human progress has always been forged. The pod at the hedge fund. The fire team on the ground. The two-pizza team in the garage. Six people in a row at a trading desk. The format isn’t new. We have just stopped building for it.
— Talib