Multi-Agent System

An arrangement of more than one AI agent, each with its own instructions, scope of action, and often its own memory and tools, coordinating to accomplish a task that no single agent among them completes alone. Distinct from a single agent that happens to call the same or different underlying models more than once under one coherent instruction set.

A setup where several AI agents coordinate, often with an orchestration layer assigning tasks to specialist agents that call tools and sometimes call each other. It introduces agent-specific failure modes (runaway loops, cross-agent cost explosions) that single-agent governance does not anticipate.

an arrangement of several specialized agents that hand work to each other (for example a researcher, a writer, and a checker); it adds division of labor but also new failure modes (cascading error, working at cross-purposes, looping) that single-agent management does not face.

Several specialized agents coordinating on a task (for example, one plans, one researches, one writes, one checks). Unlocks harder work and multiplies the ways errors and costs can compound.