Four AI agents coordinating in real time outperformed Claude Opus 4.8 on enterprise coding tasks

Researchers from Coral AI Labs and universities introduced AgentRadio, an asynchronous message-passing layer that lets AI agents communicate in real time while working through complex coding tasks. On long-horizon production repository benchmarks, four coordinated Claude Code agents nearly doubled the accuracy of independent agents and surpassed stronger single-model setups such as Claude Opus 4.8.

The results suggest that agent coordination may become as important as model size for enterprise software automation. If validated broadly, architectures like AgentRadio could shift AI coding tools toward teams of specialized agents that continuously share findings instead of relying on one increasingly large model.