The (AI) Nature of the Firm

The piece connects Coase’s insight about why humans form firms - limits on individual cognition, information aggregation, and low-level coordination that markets cannot handle - with recent developments in AI: decentralised multi-agent platforms like Moltbook and a surge of agent-focused work presented at AAAI. It contrasts current research that links a few agents or chains of tool-using bots with the broader, more complex kinds of coordination humans accomplish, arguing that existing autonomous agents are still short of the distributed, multi-role activity required for large-scale, long-horizon tasks.

Drawing on Herbert Simon’s work on administrative behavior and hierarchical information filtering, the argument asserts that the same pressures that created corporations will drive AI systems to organise into collective, firm-like structures as agent populations scale into the dozens or thousands. It forecasts a shift from standalone agents to “AI corporations” in the near future and highlights this as an opportunity for economists, organisational theorists, and mechanism designers to shape collective-intelligence architectures and decision-making mechanisms before these structures solidify.