What happened
Researchers created a system where multiple AI agents can work together using far less communication than before — success stays at 64% even when bandwidth drops to 4 bits per step, while traditional approaches collapse to 28%. This matters because real-world multi-agent systems (robots, autonomous vehicles, drones) have hard limits on how much they can talk to each other, so learning to coordinate efficiently rather than broadcasting everything changes what's physically deployable.
Why it matters
This demonstrates that multi-agent coordination doesn't require constant full communication — agents can develop shared understanding by modeling each other's positions and intentions through minimal bandwidth, which removes a major constraint on deploying swarms of robots or vehicles in bandwidth-limited environments.