针对资深开发者对全权交由AI编码代理（Agentic Development）所持的谨慎立场，Thoughtworks欧洲区首席技术官Giles Alexander指出，这种顾虑并非源于对失业的恐惧或对手写代码的偏爱，而是基于缺乏在无人监督下信任代理构建系统的经验 1。Alexander团队进行了一项为期一周的内部实验，要求由AI代理完成所有编码工作，结果显示尽管应用能够运行，但构建频繁失败，导致系统稳定性下降 1。这一现象印证了DORA 2025年的研究发现，即人工智能的采用虽然提升了交付吞吐量，却与更低的交付稳定性相关联 1。
深入分析该实验的时间分配发现，开发团队约三分之二的时间用于白板前的领域建模和边界设计，剩余三分之一的时间则专注于构建测试和检查框架（harness），这意味着架构设计和反馈机制的构建占据了大部分精力 1。Alexander强调，“构建不等于在生产环境中运行”（Built isn’t the same as working in production），并引用Charity Majors的观点进一步阐释：“预发布环境只是被美化了的笔记本电脑，只有生产环境才是真正的生产环境”（Staging is just a glorified laptop. Only prod is prod） 1。基于这些发现，文章呼吁领导者重视开发者的专业判断，以应对这一新开发范式带来的不确定性 1。
Veteran software engineers are increasingly cautious about fully autonomous AI coding agents, not due to fears of job displacement or a preference for manual programming, but because they lack trust in building systems without human oversight 1. Giles Alexander, the Chief Technology Officer for Thoughtworks Europe, highlighted this concern by emphasizing that "Built isn’t the same as working in production," suggesting that successful compilation does not guarantee operational stability in real-world environments 1. This sentiment aligns with industry perspectives cited by Alexander, such as Charity Majors’ assertion that "Staging is just a glorified laptop. Only prod is prod," underscoring the critical gap between development stages and live deployment 1.
To investigate these concerns, Alexander’s team conducted a one-week internal experiment where all coding tasks were delegated exclusively to AI agents, with zero handwritten code 1. The results revealed that the team spent approximately two-thirds of their time on whiteboard sessions focused on domain modeling and boundary design, while the remaining third was dedicated to constructing testing and check frameworks, known as harnesses 1. Although the application functioned, the project suffered from frequent build failures, leading to decreased system stability 1. These findings mirror data from the 2025 DORA report, which indicates that while AI adoption correlates with higher delivery throughput, it is simultaneously associated with lower delivery stability, prompting leaders to prioritize developer expertise when navigating this new paradigm 1.
评论
还没有评论，欢迎留下第一条。