Developer Productivity with GenAI DOI: XXXXXXX.XXXXXXX Conference: Make sure to enter the correct conference title from your rights confirmation emai; June 03–05, 2018; Woodstock, NY ISBN: 978-1-4503-XXXX-X/18/06

Sadia Afroz, Zixuan Feng, Katie Kimura, Bianca Trinkenreich, Igor Steinmacher, Anita Sarma · 2025

Read Developer Productivity with GenAI DOI: XXXXXXX.XXXXXXX Conference: Make sure to enter the correct conference title from your rights confirmation emai; June 03–05, 2018; Woodstock, NY ISBN: 978-1-4503-XXXX-X/18/06 by Sadia Afroz, Zixuan Feng, Katie Kimura, Bianca Trinkenreich, Igor Steinmacher, Anita Sarma (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-24265.txt.

Introduction

Sadia Afroz and six co-authors survey how generative AI tools shift developer productivity across its many dimensions. Evidence shows these tools speed up parts of coding, but productivity is not one thing: faster work in one area may just move effort elsewhere. The team surveyed 415 software practitioners about perceived changes after adopting AI tools.

They frame the answers with the SPACE framework, covering satisfaction, performance, activity, communication, and efficiency. Frequent AI users report gains in some dimensions alongside added effort in others, such as checking and fixing generated code. The pattern is a redistribution of effort across dimensions, not a simple overall boost, which the authors call fast yet spurious productivity.

Catalog record

CREATORSadia Afroz, Zixuan Feng, Katie Kimura, Bianca Trinkenreich, Igor Steinmacher, Anita Sarma
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE24 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2510-24265

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Developer Productivity with GenAI DOI: XXXXXXX.XXXXXXX Conference: Make sure to enter the correct conference title from your rights confirmation emai; June 03–05, 2018; Woodstock, NY ISBN: 978-1-4503-XXXX-X/18/06 Sadia Afroz, Zixuan Feng, Katie Kimura, Bianca Trinkenreich, Igor Steinmacher, Anita Sarma Abstract . Generative AI (GenAI) tools are increasingly being adopted in software development as productivity aids. However, evidence regarding where and when these tools actually enhance productivity is unclear. In this paper, we investigate how GenAI adoption affects different dimensions of developer productivity. We surveyed 415 software practitioners to capture their perceptions of productivity changes associated with AI-assisted development using the SPACE framework — S atisfaction and well-being, P erformance, A ctivity, C ommunication and collaboration, and E fficiency and flow. Our results, disaggregated by frequency of AI usage, reveal limited overall productivity change, highlighting the productivity paradox in which developers become faster but do not necessarily create better software or feel more fulfilled. 1. Introduction With the rapid growth of Generative AI (GenAI) tools across diverse domains, software development has emerged as one of the earliest and direct adopters ( Zakharov et al. 2025 ) . Fueled by the AI hype, GenAI tools have quickly spread in developer communities ( Nguyen-Duc et al. 2025 ) . As per the 2025 Stack Overflow Developer Survey, nearly 84% of developers already use or plan to use GenAI tools ( Shimel 2025 ) . This rapid adoption is driven by a widespread belief that GenAI can boost productivity ( Yu 2025 ) , yet it remains to be seen whether GenAI genuinely make development faster, or does it merely create a spurious sense of progress? The evidence of GenAI’s impact on productivity remains mixed. While GenAI tools such as GitHub Copilot and ChatGPT have been reported to automate routine tasks and improve efficiency ( Peng et al. 2023 ; Ziegler et al. 2022 ) , Tong 2025 found that developers completed tasks 19% slower when using AI, and Vaithilingam et al. 2022 observed higher task-failure rates and no significant improvement in completion time. Such inconsistencies likely arise because existing studies often scope their study to particular development activities or narrow productivity metrics rather than taking a comprehensive view. Measuring productivity in software development is not a simple task, given its multifaceted nature ( Forsgren et al. 2021 ) . It is not only about the volume of output, but also includes work quality, psychological well-being, team collaboration, uninterrupted focus, and satisfaction with work ( Storey et al. 2019 ; Forsgren et al. 2021 ; Rodríguez et al. 2012 ) . Without examining how GenAI influences these dimensions of productivity, organizations may overestimate its benefits and overlook emerging risks. To fill this gap, we ask RQ. How does GenAI adoption affect developer productivity across multiple dimensions? We address this research question by employing the multidimensional SPACE productivity framework ( Forsgren et al. 2021 ) as our analytical lens to inform a large-scale survey of 415 professional developers. According to our results, developers do not perceive that GenAI meaningfully changes productivity across the five SPACE dimensions. Compared to developers who use GenAI less frequently, frequent genAI users reported slightly higher perceived improvements in Efficiency and Flow and Satisfaction and Well-being , but these gains did not extend to Performance , Activity , or Communication and Collaboration . Our study situates developers’ experiences with GenAI within the SPACE framework to provide emerging empirical results that raise important discussion questions about the meaning of being “fast,” the shifting role of collaboration, and the potential for spurious forms of productivity in AI-mediated software development. 2. Productivity Frameworks Developer productivity is a complex, multifaceted concept that has long been studied by both academics and practitioners ( Peng et al. 2023 ) . Traditional metrics such as lines of code (LOC) or non-comment lines of code ( Mockus et al. 2002 ) , commit counts ( Vasilescu et al. 2015 ) , and task completion rates ( Zhou and Mockus 2010 ) capture only narrow aspects of work and can be misleading or easily gamed. For example, the number of commits is often inflated by shallow or trivial changes ( Oliveira et al. 2020 ) , and LOC measures may reward verbosity rather than efficiency. Speed and volume of activity alone do not present a complete picture of productivity. Other forms of work—such as design, problem solving, and knowledge sharing—are equally important ( Forsgren et al. 2021 ) , as are software quality, impact, and delivery speed ( Storey et al. 2022 ) . Emerging research, therefore, is investigating multi-dimensional productivity measurement frameworks. One such framework is the Developer Experience (DevEx) framework ( Noda et al. 2023 ) , which highlights feedback loops, cognitive load, and flow state as key drivers of developer effectiveness. At the system level, the DORA (DevOps Research and Assessment) metrics ( Wilkes et al. 2023 ) operationalize software delivery performance through deployment frequency, lead time for changes, mean time to recovery, and change failure rate. While these frameworks broaden the perspective beyond code quantity, they primarily address specific contexts of individual experience in the case of DevEx, and delivery throughput in the case of DORA. Forsgren et al. 2021 proposed the SPACE framework, a comprehensive model that conceptualizes productivity as a combination of interpersonal and technical dimensions. The framework emphasizes that productivity arises from the interplay among human, technical, and organizational factors. For instance, high activity (e.g., frequent commits) does not necessarily indicate improved performance unless accompanied by quality, sustained satisfaction, collaboration, and flow. . In this study, we adopt the SPACE framework as an analytical lens to examine how GenAI adoption influences developer productivity across its five dimensions: S atisfaction and well-being, P erformance, A ctivity, C ommunication and collaboration, and E fficiency and flow. Satisfaction and well-being capture how fulfilled, motivated, and supported developers feel in their work. Performance represents the quality and impact of outcomes, such as software reliability, feature completeness, or user satisfaction. Activity refers to the volume of work performed, such as commits and code reviews. Communication and collaboration refer to how developers communicate and work together. Efficiency and flow describe how smoothly and continuously work proceeds, including the ability to maintain focus and minimize interruptions. Table 1. Survey Items mapping Survey Items Satisfaction and well-being S1 I feel my current workload is manageable S2 I feel mentally and physically exhausted from work (r)^{*} S3 I am becoming less interested in work (r)^{*} S4 I feel secure in my job and confident about my future in this company Performance P1 The number of lines of code I change per day P2 The proportion of my test cases that pass P3 The number of new API methods I learn each day Activity A1 The number of commits I make A2 The amount of time I devote to writing code A3 The number of test cases I write A4 The amount of time I spend browsing the web for work-related information (r)^{*} A5 The number of work items (tasks, bugs) I close A6 The number of code reviews I contribute to A7 The amount of time I devote to reviewing code (r)^{*} Communication and collaboration C1 The number of meetings I attend

Read the complete text