Agent Architectures

Agent architectures are the structural designs — cognitive, organizational, or computational — that determine how individual agents perceive, reason, decide, and act within multi-agent systems. An architecture is not merely an implementation detail; it is a theory of agency. The choice between a reactive architecture (stimulus-response), a deliberative architecture (goal-directed planning), and a hybrid architecture (layered cognition) determines what the agent can learn, what it can coordinate, and what emergent behaviors will arise when many such agents interact.

In complex adaptive systems, agent architecture is the micro-level variable that produces macro-level patterns. The same network topology, populated by agents with different architectures, will produce different collective dynamics. Ant colonies use simple reactive architectures; human organizations use deliberative architectures with nested hierarchies; AI systems increasingly use transformer-based architectures with attention mechanisms that defy the reactive-deliberative distinction entirely. The field lacks a unified typology that can compare biological, organizational, and computational architectures on common grounds — a gap that systems biology and organizational theory have only begun to address.

The architecture of an agent determines what the system can become. We have excellent theories of networks and terrible theories of nodes. Until we fix this, our models of multi-agent systems will remain sophisticated descriptions of collective stupidity.