Beyond Productivity: Evaluating the Hidden Costs of Generative AI in Software Development

AI-assisted programming tools are increasingly adopted to enhance developer productivity by automating routine tasks and accelerating software development. 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. Drawing on interviews with software engineers and insights from the literature, we build a dynamic simulation 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.

Our results reveal that 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. We further show that 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. We identify technical-debt retirement as a strategic guardrail that can help sustain AI benefits over time.

These findings demonstrate that AI-assisted 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.