AgentSWE：编码智能体能否构建你真正想要的智能体？
原标题：Paper page - AgentSWE: Can Coding Agents Build the Agent You Actually Want?
AI 摘要
该论文提出 AgentSWE，一个面向智能体软件工程（agent software engineering）的基准测试，覆盖智能体开发生命周期的三个阶段：从自然语言需求创建完整智能体、编辑生产级智能体代码库、以及面向可量化目标的优化。其 25 个任务被设计为“委托”而非“考试”，构建者只能针对少量示例迭代，而验收则在覆盖完整需求的留出用例上以硬性通过/失败门槛评分。作者评估了多种模型与框架下的编码智能体，发现没有一个能接近满足需求，构建者演示效果与实际交付效果之间的差距是常态而非特例，并总结了过拟合可见用例、代码看似完整但运行时空洞、以及无依据的完成声明等反复出现的失败模式。
正文节选
AgentSWE: Can Coding Agents Build the Agent You Actually Want? Abstract As the production of software passes into the hands of software itself, coding agents are increasingly commissioned to build agents fitted to their users' own workflows. This delegation carries a quiet assumption, that the delivered agent will faithfully serve the requirement it was built for, rather than merely appear to. No existing benchmark tests that assumption. We introduce AgentSWE, a benchmark for agent software engi