Abstract

Abstract for: Generative AI and Software Quality Dynamics

Introduction AI-assisted programming tools are increasingly adopted to enhance developer productivity by automating routine tasks and supporting activities such as code generation, debugging, and documentation. While early evidence highlights short-term productivity gains, our study shows that excessive reliance on AI can unintentionally accelerate the accumulation of technical debt, leading to escalating maintenance costs and eventual revenue erosion.
Approach Drawing on interviews with software engineers and insights from the literature, we build a system dynamics model to examine the long-term impacts of AI-assisted coding under varying conditions of project horizon, legacy-system interaction, developer skill, and market dynamism.
Results Productivity benefits follow an inverted-U pattern: there exists a “sweet spot” where AI assistance improves overall performance, but beyond this threshold, compounded technical debt undermines system maintainability and profitability. Greenfield environments and fast-evolving markets can tolerate higher levels of AI assistance, whereas brownfield (legacy) settings and teams with less skilled developers are substantially more susceptible to long-term decline. Technical-debt retirement acts as a strategic guardrail that can help sustain AI benefits over time.
Discussion Our findings demonstrate that AI-assisted software development is not simply a choice of productivity tool, but a strategic decision that must consider programming environment constraints, system longevity, skill composition, and market factors.

Introduction	AI-assisted programming tools are increasingly adopted to enhance developer productivity by automating routine tasks and supporting activities such as code generation, debugging, and documentation. While early evidence highlights short-term productivity gains, our study shows that excessive reliance on AI can unintentionally accelerate the accumulation of technical debt, leading to escalating maintenance costs and eventual revenue erosion.
Approach	Drawing on interviews with software engineers and insights from the literature, we build a system dynamics model to examine the long-term impacts of AI-assisted coding under varying conditions of project horizon, legacy-system interaction, developer skill, and market dynamism.
Results	Productivity benefits follow an inverted-U pattern: there exists a “sweet spot” where AI assistance improves overall performance, but beyond this threshold, compounded technical debt undermines system maintainability and profitability. Greenfield environments and fast-evolving markets can tolerate higher levels of AI assistance, whereas brownfield (legacy) settings and teams with less skilled developers are substantially more susceptible to long-term decline. Technical-debt retirement acts as a strategic guardrail that can help sustain AI benefits over time.
Discussion	Our findings demonstrate that AI-assisted software development is not simply a choice of productivity tool, but a strategic decision that must consider programming environment constraints, system longevity, skill composition, and market factors.