研究揭示 AI 编程工具未能提升交付速度
一项基于 30 次访谈与多篇论文的研究指出，AI 编程工具显著提升了写代码环节的效率，但并未同步转化为更快的软件交付，形成“活动—产出落差”。作者认为瓶颈在于评审、集成、测试与协作等下游环节，而非编码本身。
报道正文
本文目录3 个章节
Talk · UC Irvine software engineering seminar · October 30, 2026
Why aren’t we shipping faster now?
AI, activity, and outcomes in software engineering
21 minutes, with captions. Also on YouTube.
Slides
- Handout19 pages, 3.8 MB One page per picture. The version to read.
- Full deck41 pages, 4.0 MB Every build step, as shown in the talk.
Papers
Cha, Badea, Baym, Bird, Butler, Murphy-Hill, Hadley. The Activity-Outcome Gap: Why AI Does Not Solve Software Engineering. In submission, 2026. The 30 interviews; a link will appear here when the paper is public.
Demirer, Musolff, Yang. Writing Code vs. Shipping Code: Productivity Effects Across Generations of AI Coding Tools. NBER Working Paper 35275, 2026. nber.org/papers/w35275
Choudhuri, Butler, Bird, Badea, Houck, Sarma. “To Boldly Go Where No One Has Gone Before”: What Has AI Let Developers Unlock? 2026. arXiv.23708
Jensen Huang, Computex keynote, Taipei, 29 May 2023, as reported by Reuters.
Yang, He, Cheng, Butler, Murphy-Hill, Badea, Bird. (When) Should a Software Factory Be Lights-Off? Benchmarking Code Review Agents on Bug Detection. 2026. arXiv.13668
Choudhuri, Bird, Badea, Gerosa, Sarma. You Shall Not Pass! Where and Why Developers Draw the Line on AI Autonomy. 2026. arXiv.00533
Miller, Choudhuri, Ulloa, Haniyur, DeLine, Storey, Murphy-Hill, Bird, Butler. “Maybe We Need Some More Examples”: Individual and Team Drivers of Developer GenAI Tool Use. ICSE 2026. arXiv.21280