AgentRadio: Passive Awareness for Long-Horizon Multi-Agent Collaboration

2026-07-30

AI Summary

This work addresses the efficiency bottlenecks faced by large language model agents in complex code comprehension tasks that require cross-file tracing, build execution, and evidence integration. To overcome the limitations of conventional systems where communication and execution are mutually exclusive, the authors propose an asynchronous multi-agent collaboration framework. This framework employs a communication layer built upon a triadic primitive of threads, messages, and mention-waiting, augmented with a passive awareness mechanism that enables agents to dynamically receive and incorporate information from teammates without interrupting foreground tasks. Coupled with a five-phase role negotiation protocol, the approach achieves a 62.1% task completion rate on the SWE-Atlas QnA benchmark using a four-agent system—representing a 29.8 percentage point improvement over single-agent baselines—with performance gains markedly increasing as task difficulty rises.

Abstract

Understanding large codebases is a long-horizon task for Large Language Model (LLM) agents: answering a single question can require building and running the software, tracing execution across files, and synthesizing evidence over tens of minutes. On SWE-Atlas QnA, a benchmark of long-horizon questions over production repositories, a single Claude Code agent (Opus 4.6) resolves only 32.3% of tasks. Dividing the work among agents with clean contexts mitigates this limitation. However, the subtasks of code comprehension are interdependent. One agent's findings can rewrite another's task, so agents must coordinate during execution, not only at phase boundaries. Existing multi-agent systems support such exchange only between phases, through staged handoffs or synchronized rounds. Communication and work remain mutually exclusive. A discovery made mid-execution cannot be shared until the next boundary. We present AgentRadio, an asynchronous message-passing layer that equips coding-agent harnesses with three primitives: threads, messages, and waiting for mentions. The last runs as a background task, surfacing teammates' messages without interrupting foreground work, so each agent remains passively aware of its peers and folds new findings into its ongoing task. Under a five-phase protocol of division of labor and negotiation, four agents organized by AgentRadio resolve 62.1% of tasks, 29.8 points above a single agent and above Claude Code with the newer Opus 4.8 (57.2%). Rubric-level analysis shows the gain growing with task difficulty, consistent with mid-course correction as the underlying mechanism. Our code is available at https://github.com/Coral-Protocol/AgentRadio.

Problem

• long-horizon collaboration
• multi-agent coordination
• asynchronous communication
• code comprehension
• passive awareness

Innovation

• AgentRadio
• passive awareness
• asynchronous message-passing
• multi-agent collaboration
• long-horizon reasoning