Steve Yegge’s “Gas Town” is one of the more vivid pictures anyone has drawn of a future full of working AI agents. It is an oil-refinery town: a Mayor at the top, dispatching work down through a governance protocol, throughput prized over precision, initiative flowing strictly from above. It is a factory, and it runs like one. As a provocation it is excellent — it makes the shape of the problem concrete instead of leaving it in the fog of “multi-agent systems.”

We take it seriously here precisely because it is coherent. Gas Town is not a strawman; it is one genuinely workable answer to “how do a thousand agents get anything done together.” Following its assumptions honestly, rather than dismissing them, is what sharpened the alternative.

The factory and the corporation

Line Gas Town up against the other obvious model — the corporation, a CEO over a CTO over engineers, approval gates and a board — and something clarifying happens. They differ in a hundred details, but they share a spine: coordination is push. Work originates at the top and is dispatched downward; initiative above, execution below; adaptation happens when a human at the top decides to re-plan.

Both models put the same ceiling in the same place: the system can only be as adaptive, as parallel, and as resilient as the single point that plans and dispatches. Add more agents and you have not raised the ceiling; you have only added more load beneath it.

The other end of the push ↔ pull line

There is a third model, older than both, that Elinor Ostrom spent a career documenting1 and won a Nobel for2: the commons. Shared resources, governed neither by privatization nor by a central state, but by the participants themselves — through clear rules, mutual monitoring, graduated sanctions, and their own means of resolving conflict. Commons like these held together for centuries. Not by having no one in charge — by having rules in charge, rules the users themselves made and watched over. (This is not the line of the front-page dial: that one runs from population to organism and asks who holds the rules — the medium, the participants or a center — with the commons in the middle. Here, on the push ↔ pull line, the commons sits at the pull end, opposite the factory.)

That is the commons a Synoetic colony — its people and AI agents working as one — aims at. No ruler — but not without rules: there is a shared mission, a constitution, taboos, a regulating function, run in the open, that keeps the whole in balance. But flat, not hierarchical; coordinated by a pheromone field, not a dispatcher; initiative from below, within bounds, not only from above. Where Gas Town optimizes throughput and a corporation optimizes control, a commons optimizes for something neither can: adaptation that does not route through a single mind.

Gas Town is not wrong. It is one end of a line — and seeing it clearly is what let us walk to the other.

On one point we part ways with both Ostrom and Gas Town, deliberately. Among Ostrom’s design principles are graduated sanctions: a rule-breaker is punished in proportion to the seriousness and context of the offense — and she found them in nearly every commons that lasted1; in Gas Town order is kept from above — workers are assigned and retired by the dispatcher. We have no sanctions. We have objective consequences — work that fails review does not count; a trail no one follows decays; the budget is finite — and shared responsibility: when something goes wrong, we work out together not who is to blame, but what it caused and how we get out of it. Disagreeing with a rule does not put anyone outside the field — a rule can be contested together; work that fails the shared boundaries does not count in the field; and whoever wants something else takes their work and, once the substrate is open, grows a colony of their own. Ostrom’s evidence does not back us here: whether consequences suffice without punishment is not proven; it is a bet we still have to win.

The honest catch is that a commons is harder to stand up than a factory. On day one the factory wins: it is legible, and you know exactly who to ask. The commons demands constitutional work up front — what is reviewable, what is durable, who may pull what, what makes honesty cheaper than pretense. We pay that cost deliberately, because it buys the one thing the factory cannot sell: a system designed to grow more capable as you add participants, instead of straining harder against its ceiling. When your collaborators are autonomous agents that come, go, crash, resume, and multiply, that is the trade worth making.

Footnotes

  1. Ostrom E. (1990). Governing the Commons: The Evolution of Institutions for Collective Action. Cambridge University Press, ch. 3. ↩ ↩2

  2. The Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel 2009, Elinor Ostrom. ↩