What is Multi-agent system?

A multi-agent system is one where more than one AI agent collaborates to accomplish a goal, instead of a single agent doing everything in one loop. The motivation is specialization and parallelism: a researcher agent gathers sources while a writer agent drafts, a planner decomposes a request and worker agents tackle the pieces, or a reviewer agent checks another's output. A common architecture is the supervisor-worker pattern, where one orchestrating agent delegates subtasks and synthesizes the results. Multi-agent systems introduce coordination problems that a single agent never faces: who does what, how intermediate results flow between agents, how to avoid duplicated or contradictory work, and how to keep cost bounded, which is the domain of agent orchestration. The Model Context Protocol helps by giving every agent in the system a uniform way to reach the same external tools, and shared memory is what keeps them coherent: when all the agents read and write a common memory store, a fact one agent learns is immediately available to the others, so they operate on a consistent picture of the task rather than drifting apart. Without shared context and tool access, the agents in a multi-agent system tend to fragment.