The Superorganism, in Six Diagrams

Many minds, one dense text. Conway's law: the code's internal link graph takes the shape of the org that wrote it. Superintelligence already existed as the well-run organization; access was gated to the few who mastered onboarding, documentation, promotion, and culture.

Zoom into any node of the org graph and you find a whole organization: one human, day-one tech lead, commanding an agent legion that can grow tenfold for an afternoon. The genuinely new problem rides along: agents produce faster than their own human can verify, so trust now lives inside the node.

One set of media, two kinds of reader. Humans get rich text and interfaces; agents grew up on the same substrate, because the pretraining corpus is the web. HTML is the deep dual-native case. New agents arrive already speaking repo, and documentation becomes the interface to half the workforce.

The humanistic controller, fastest loop inside. Loop one: the bench, reflex-speed review in the live session. Loop two: the responsible party, one named human per deliverable, a Markov blanket around agents spawning agents. Loop three: the meta-RP's social world model, Dunbar-bounded near 150 people, 30 to 60 of them active.

Dunbar's number is a hard ceiling, so the coordinator role copies downward instead. The trust ratchet raises two bars per person, complexity trusted and scope owned. Members of technical staff clear the top bar, and over nine to sixteen months start webs of their own. Branching factor around thirty; capability compounds while no single mind holds more than its share.

The objective function, drawn as two network designs. Left: strong nodes everywhere, redundant links, any node can fail and the network survives. Right: the bunker civilization, one hub and managed dependents, a single point of failure in an era of pocket superintelligence. The fractal org is the left picture, built small first.