The Impact of Generative AI on Agile Software Development: A System Dynamics Approach

Abstract

Generative Artificial Intelligence (GenAI) has demonstrated remarkable potential to boost the productivity of software developers. However, its overarching impact on agile software development projects remain largely unknown. Given agile's collaborative and iterative nature, GenAI adoption in such environments may yield unintended consequences such as developer overdependence and excessive technical debt accumulation. To investigate these effects, this study employs a system dynamics (SD) approach to model how GenAI-driven code, test, and documentation generation impacts project performance (software quality and development productivity) and human resource (developer knowledge and skills) outcomes over time. An SD model of 'GenAI-augmented agile software development' is developed and validated using qualitative and quantitative data from software professionals and real-world agile projects. The model reveals reinforcing and balancing feedback loops triggered by rapid code generation, refinement of generated code, knowledge and skill loss, tech-debt accumulation, and refactoring activities. It was found that while GenAI initially accelerates individual developers' coding productivity, inadequate comprehension and refinement of generated code can degrade software quality in subsequent sprints, leading to substantial rework and refactoring effort that throttles developers' capacity for new feature development. However, disciplined refinement practices and strategic GenAI usage for testing and documentation mitigate these drawbacks, safeguarding software quality and sustaining productivity over the course of the project. These findings contribute to the growing literature on GenAI's role in software development and offer actionable guidance for practitioners and organizations seeking to integrate GenAI effectively into agile software projects.