Abstract
Research on artificial intelligence agents commonly evaluates the capabilities, incentives, and interactions of individual agents. Yet increasingly consequential systems are not individual agents but collectives: ensembles of language-model agents, hybrid human–AI teams, and organizational arrangements in which observations, recommendations, decisions, and actions are distributed across people and synthetic components. Strong constituent agents do not necessarily produce a coherent, adaptive, or accountable collective. This article develops the Viable System Model (VSM) as a diagnostic framework for that organizational problem. It distinguishes AI agents, agent-based models, and deployed agentic systems; defines collective viability as the continued maintenance of a distinct organization under changing conditions; and translates the VSM’s functions of operation, coordination, regulation, intelligence, and identity into research problems for hybrid human–synthetic systems. The proposed mapping is functional rather than ontological: it neither assumes that every agent collective instantiates the VSM nor assigns one agent to each VSM function. The synthesis identifies six hard problems: recursive boundary formation, collective identity, channel capacity, conflict mediation, exception handling, and distributed accountability. From these it formulates falsifiable propositions and demonstrator designs. The resulting agenda shifts evaluation from isolated agent capability toward the organizational conditions under which populations of agents can remain effective, adaptive, governable, and answerable.
Showing the abstract — retrieve the full paper via the Exa API.