Work

Build Fast, Fix Later: Downstream Consequences of AI-First Practice in Software Development

Type: article
Abstract: Software development is increasingly organized around human-AI collaboration, with AI coding assistants (AICAs) becoming embedded in everyday work. Existing research explains AICA adoption, productivity benefits, and risks, but does not fully explain how this human-AI way of working becomes embedded in practice or how its consequences unfold across levels over time. Drawing on a practice perspective, we examine how AI-first practice emerges in software development and generates downstream consequences. Based on a grounded theory study of AICA use in a software-development firm, we find that rising delivery expectations and the ready availability of AICAs make it more likely that developers turn to AI as a first-line resource for understanding and progressing software development work. As these processes become recurrent, they reveal a temporal asymmetry: immediate gains in speed and throughput mask later problems in developer capability and knowledge, codebase governability, and organizational oversight of software work.
Language: English
Open Access status: green
OpenAlex ID: W7222910825
Cites: 24
Cited by:
Related to: 10