Multi-Agent System

A multi-agent system splits a job across several agents, each with its own role, prompt, and tools. A classic example is a researcher agent that gathers facts, a writer agent that drafts, and a reviewer agent that checks the result. The idea is that focused agents with narrower context do each part better than one agent juggling everything. The trade-off is cost and complexity. Every agent is more model calls and more tokens, and coordinating them adds failure modes: agents can loop, disagree, or pass along bad context. That is why the common advice in 2026 is to start with a single agent and only split into multiple agents when a clear division of labor justifies the overhead. Frameworks like CrewAI, AutoGen, LangGraph, and Semantic Kernel all support multi-agent setups, but they coordinate differently. CrewAI uses roles and tasks, AutoGen uses conversations between agents, and LangGraph uses an explicit graph. The right model depends on whether your problem is a structured pipeline or an open-ended collaboration.