跳到正文
arXiv cs.AI· Annemarie Wittig, Alina Mailach, Norbert Siegmund·· 4 小时前AI 评分38

系统综述:AI 编程助手对开发者生产力的证据碎片化、结论不一

Fragmented Markers, Mixed Results: A Systematic Review of AI Coding Assistants and Developer Productivity

AI 导读

一项分阶段系统文献综述梳理 AI 编程助手对开发者生产力的影响,共识别出 187 个生产力及生产力相关指标,而单项研究通常只使用 5 个,证据基础广泛但高度碎片化。明确以生产力为主题的研究以时间、产出和制品类代理指标为主,纳入 SPACE 框架后满意度、心理层面与幸福感才明显可见。报告效应多为正向,但这一倾向在主观感知类发现中强于客观或经统计检验的发现,现有证据尚不足以支撑简单、脱离情境的结论。

正文 · 原文

View PDF HTML (experimental)

Abstract:AI coding assistants are increasingly embedded in software engineering practice, yet the literature still offers no cumulative, easily interpretable picture of their effects on software developer productivity. The central problem is not a lack of studies, but a lack of comparability: prior work operationalizes productivity through heterogeneous markers, settings, and reporting styles, making cross-study conclusions difficult. To address this, we conduct a staged systematic literature review. In the first stage, we synthesize studies that explicitly examine developer productivity. In the second, we extend the scope to productivity-related developer outcomes using SPACE as a theory-informed lens, capturing human-centered evidence that explicit productivity studies often omit. Across both stages, we analyze how productivity is operationalized, which confounding factors are discussed, and how reported effects are distributed.
We identify 187 unique productivity and productivity-related markers, while individual studies typically only use five, revealing a broad but highly fragmented evidence base. Explicit productivity research is dominated by time-, output-, and artifact-oriented proxies, whereas the broader SPACE-aligned corpus makes satisfaction, psychological aspects, and well-being substantially more visible. Reported effects are often positive, but this pattern is stronger in perceived and tendency-based findings than in objective or statistically evaluated ones. Overall, the literature suggests that AI coding assistants can support productivity, but current evidence does not yet support simple, context-independent conclusions. The key contribution of this review is therefore a structured basis for making productivity evidence more interpretable, multidimensional, and comparable across studies.
Comments: Accepted at the 41st IEEE/ACM International Conference on Automated Software Engineering (ASE '26)
Subjects: Software Engineering (cs.SE); Artificial Intelligence (cs.AI)
Cite as: arXiv:2611.21632 [cs.SE]
  (or arXiv:2611.21632v1 [cs.SE] for this version)
  https://doi.org/10.48550/arXiv.2611.21632

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Annemarie Wittig [view email]
[v1] Wed, 16 Sep 2026 13:40:40 UTC (74 KB)

来源:arXiv cs.AI · arxiv.org