More Agents Is Not All You Need
A Google Research paper tested 180 agent configurations and found multi-agent systems average -3.5% performance. I tested it on code generation and got a different result.
Everyone building AI agent systems right now assumes the same thing: more agents, better results. Split the work across specialized agents, let them collaborate, and the output improves. A research paper from Google Research, DeepMind, and MIT tested that assumption across 180 controlled configurati...More ›