📄 中文摘要
大多数AI编码代理通常独立工作，接收问题后自行解决并提供修复方案。然而，Agyn的研究团队提出了一个新思路：使用编码代理团队，赋予其真实角色、审查流程和协调机制。研究表明，这种团队协作模式在解决问题时更有效，能够更好地模拟真实软件开发中的协作过程。通过这种方式，问题的研究、修复和审查环节得以有效整合，最终提升了整体的解决效率和质量。
编码代理团队的表现优于单一代理：在SWE-bench验证中达到72.2%
📄 English Summary
Coding Agent Teams Outperform Solo Agents: 72.2% on SWE-bench Verified
Most AI coding agents operate independently, tackling issues and providing fixes on their own. However, a research team from Agyn proposed a different approach: utilizing a coding agent team with defined roles, review loops, and coordination. The results indicate that this collaborative model is more effective in problem resolution, closely mirroring the actual collaborative processes in software development. By integrating research, implementation, and review stages, the team-based approach enhances overall efficiency and quality in addressing coding challenges.