编码代理如何优化软件并报告性能验证？大规模开源PR实证研究
原标题：How Do Coding Agents Optimize Software and Report Performance Validation? A Large-Scale Empirical Study of Open-Source Pull Requests
AI 摘要
该研究对截至2026年6月的1130个AI编码代理提交的性能优化PR与1130个人类提交的PR进行大规模实证对比。结果显示代理PR合并率更低（54.4% vs 73.3%），但优化策略和验证报告率相近；代理更依赖静态推理而非基准测试（46.1% vs 57.2%），且约半数已验证PR未报告量化性能指标，权衡也极少被量化。研究指出代理优化实践正趋近人类，但支撑证据仍不完整，呼吁建立衡量预期效果与成本的评估基础设施和审查标准。
正文节选
How Do Coding Agents Optimize Software and Report Performance Validation? A Large-Scale Empirical Study of Open-Source Pull Requests Abstract Software performance optimization requires identifying bottlenecks and improvement opportunities, and empirically evaluating the effects and trade-offs across performance objectives. Autonomous AI coding agents submit performance-oriented pull requests (PRs) to open-source projects, yet it remains unclear whether these changes reflect established optimizat