Multi-Agent System
An architecture in which multiple AI agents, each with distinct roles and capabilities, collaborate or compete to accomplish tasks that exceed what any single agent could do within one context window.
Multi-agent systems decompose complex work along two dimensions: specialization (a research agent, a code agent, a reviewer agent each optimized for one domain) and parallelism (multiple agents working simultaneously on independent sub-tasks). This mirrors how human organizations are structured, and the same coordination challenges apply.
Communication between agents happens through shared message queues, structured handoff payloads, or a shared workspace (like an Obsidian vault or a database). Governance requires clear ownership—each task has exactly one responsible agent at any moment—to prevent both deadlock and duplicate work.
As of 2026, frameworks like Paperclip, LangGraph, and the Claude Agent SDK provide primitives for spawning sub-agents, routing messages, and enforcing budget limits across an entire fleet. Key open problems include trust between agents (can agent A trust agent B's tool results?) and emergent behavior in large swarms.