Four AI Agents Beat Claude Opus 4.8 Via Teamwork

Researchers at Coral AI Labs and multiple universities developed AgentRadio, an asynchronous message-passing system that lets AI agents communicate in real time between execution steps. This solves a key flaw in traditional multi-agent systems, which lack mid-task coordination. Tested on long-horizon enterprise coding benchmarks, four Claude Code agents using AgentRadio nearly doubled task accuracy compared to Claude Opus 4.8 alone. The architecture allows agents to course-correct on interdependent subtasks rather than pursuing dead ends until a formal review.