The multi-agent hype is largely misunderstood. The idea that you can split a complex problem into peer agents that gossip and coordinate — and expect it to work — is not where current model capabilities actually are.