"Multi-Agent" Is Often a Single Agent: What 86 Repos Actually Implement
Every agent framework README says "multi-agent". AutoGen, CrewAI, LangGraph, MetaGPT each have 20k+ GitHub stars, and the term "multi-agent" appears in thousands of repo descriptions. But what do projects that call themselves multi-agent actually implement? Until now, nobody measured the population — only how to build frameworks, or theory about whether multi-agent is "just prompting". A new…
A recent study analyzed 86 self-described multi-agent repositories and found that 68.2% of them actually implement a single model instance or a non-agent system. The term "multi-agent" appears in the descriptions of thousands of repositories, but the actual implementations vary greatly. Orchestrator-worker topology was the most common among genuine multi-agent systems, while judge/critic agents were rare.
The census also revealed that many repositories with the multi-agent label use framework dependencies without claiming the label. The research highlights a gap between the label and reality, indicating that most self-described multi-agent systems are not truly multi-agent systems. The study emphasizes the importance of properly defining and implementing multi-agent systems.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
