Cognition

Anthropic tested 45 AI agents collaborating: bigger teams coordinated worse

Anthropic's 45-agent bug-hunting swarm found 266 flaws using 27M tokens — but the real story is coordination failure. Stronger models ≠ better coordin