Coordination Without a Pyramid

Over the past weeks I have been adding links to a chain. Risk is what we can measure and model, and it changes with AI. Trust is the leap we take across what our models of risk cannot measure yet. Autonomy is how we take advantage of that trust. We either delegate, or we recognize that we can work with someone as a peer. The next link follows from these three: when many autonomous actors exist, how do they move together?

What the Layers Were For

For most of history the answer was a hierarchy, and it emerged naturally and necessarily from the kind of autonomy and coordination that was possible at the time. Objectives, or to put it less politely, orders, flowed down the chain of command. Reports went up, confirming that the objectives had been reached. Questions and clarifications travelled in both directions. Each layer in between existed to translate, check and relay what it received from the layer next to it.

Today any actor, and by actor I mean any participant, human or AI agent, can set its own goals and execute them. Once the actor at the bottom can do that, most of the vertical traffic has nothing left to carry.

The Price of Trust Between Peers

In 1937 Ronald Coase asked why firms exist at all, if markets are so efficient. His answer was that coordinating exclusively through the market is expensive. Finding a counterpart, building trust, writing the agreement and checking that it is fulfilled each cost time and money. A hierarchical organization was cheaper than a network of peers who had to establish trust with each other one transaction at a time.

Blockchain-based networks promised to change that price, and we seldom discuss them now, because in a bear market they are less fashionable than they used to be. Everyone believes they can ignore them until they wake up and realize that they should have acted sooner. New ways of coordinating are still among their main advantages.

AI lowers the price of peer trust from another direction. Context is shared instantly. Work is legible and transparent. Verification is cheap, as I argued previously, and agreements can be drafted and checked in minutes. When horizontal trust gets cheap, the boundary between the organization and the network of peers moves. I expect to see AI and blockchain combined for exactly this purpose, and in many ways it is already happening.

Toyota Did It First

The technology makes this kind of coordination accessible to many more people, but the idea is not new. Toyota has worked this way for decades. Through hoshin kanri, leadership sets a few strategic directions, and the teams decide how to reach them, negotiating the targets back and forth, with the levels above them and also with their neighbors. In the most famous version of the story, anyone on the production line can stop it on seeing a defect, because the person nearest the problem has the best information.

What Toyota kept at the top were three kinds of decisions: how to formulate the mission, where the limits to autonomy lie, and how to allocate what is scarce, whether money, materials or anything else that is not, or not yet, abundant.

Reporting Becomes Visibility

My fleet of agents and my team work like this. Everything we discuss in a recorded meeting, and all the code, comments and documentation the agents produce, goes into a knowledge base that is visible to everyone. Each session writes its own record, and humans and agents appear side by side as line items in the same daily account.

As a consequence, nobody’s job is to summarize and report upward, because anyone can look. When we do meet, it is to brainstorm, to surface roadblocks and to ask for help. The status meeting, where people take turns reading out updates, has no reason to exist.

The same logic extends beyond the organization. For a year and a half, visitors to my website could chat with a bot that answers in my name. That forces a human to spend precious time extracting information, which they either forget or paste back into their own agent because it mattered enough to keep. It is better for the visitor’s agent to ask the questions it was told are relevant, or that it decides are relevant, and keep the answers. Members of the Ikigai Collective have just built exactly this for the Collective’s website, so that anyone’s agent can search and read it directly. When tens of billions of agents interact, we cannot rely on humans to talk to each of them. Humans should stop being the slowest and least reliable component in the chain of information.

Arbitration Does Not Get Cheaper

A bottleneck is never eliminated. It only moves somewhere else, and in this case it moves to arbitration. When two initiatives, each self-directed, want the same budget, the same customer or the same week, someone has to decide.

In the early 1970s, Jo Freeman wrote The Tyranny of Structurelessness. Groups that claim to have no structure, she argued, grow a hidden one, and pretending it is not there makes it harder to displace and harder to challenge. A lightweight structure that everyone can see is much better than a secret pyramid that survives, or gets rebuilt, out of sight.

The Shape

The shape of these new organizations is not clear yet. It is certainly not a steep pyramid with many layers, and a completely flat line is a utopia. Jack Dorsey, the founder of Block, said that he is eliminating middle management and expects that within a couple of years everyone will report to the CEO. How that will work in a company with thousands of people remains to be seen.

What we can describe are the parts. A small spine holds the mission, the boundaries set by the limits of autonomy, and the function of arbitration. Around it, cells form opportunistically, combine and dissolve as needed. Networks of microenterprises have worked this way for years. What is new is that one human with a fleet of agents can be a complete and effective cell. This is the single-person millicorn seen from the side of the organization that works with it. Anthropic has just announced that it acquired Mithril Medicine, a one-person company developing treatments for two rare forms of muscular dystrophy. Its founder, Cody Tranbarger, brought the appropriate obsession and used Claude to coordinate contracts, logistics and experiments with eleven external labs. The cell became part of a much larger structure that is building vertical AI for biology and health.

Who Decides

That leaves the question I most want to see answered. In a structure of autonomous peers, when two of us want the same scarce thing, who decides? If the answer is whoever the group defers to by default, the hierarchy is still there, only hidden. We can make the mechanism explicit, or we can try to design conflict resolution that is coevolutionary and not adversarial. I would like to see the coordination of resource allocation treated as an engineering problem, open to the same kind of invention we are applying to everything else.