Answers are grounded in this paper and cite their sources.
Multi-agent AI systems are commonly organized by importing human institutional forms: teams, managers, committees, channels, voting, markets, and role hierarchies. This paper asks a different question: if machine agents are allowed to exploit their own substrate properties rather than inherit human organizational priors, what institutional forms should emerge, and under what conditions? We use Motif-Guided Pressure Evolution (MGPE) as a disciplined derivation procedure. The central recurrence is pressure Pᵢ → structural mutation Δᵢ → affordance unlock Uᵢ → successor pressure Pᵢ₊₁. We identify five control variables — verifiability, reversibility, error correlation, representation compatibility, and consequence accumulation — and derive candidate phase boundaries for seven recurring institutional morphologies: clone swarms, oracle stars, polycrystal archipelagos, membrane cells, branch lattices, bonded principal trees, and clocked strata. The theory predicts, among other results, that homogeneous replication saturates at a ceiling set by the distribution of per-item competence rather than by agent count; commit gating emerges when expected irreversible loss exceeds verification cost; representation incompatibility produces cluster boundaries; persistent identity reappears where consequences must accumulate; and institutional self-modification becomes unstable when reorganization occurs faster than causal effects can be identified. Two predictions arise only from the interaction of laws: the value of agent diversity rises with access to verification, and rollback infrastructure raises the ceiling on safe adaptation only until the identification bound binds. We distinguish coarse phase selectors from internal topology modifiers and show why mature systems should be nested composites rather than members of a single class. We close with a blind-retrodiction protocol and a set of sharp exclusions intended to make the framework falsifiable rather than merely descriptive. Working paper — theory and test protocol, not yet empirically validated.
No takes yet. Share an insight, caveat, or question.
Nikit Phadke (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: