Debt Behind the AI Boom: A Large-Scale Study of AI-Generated Code in the Wild (arxiv.org)

A large-scale study published on the impact of AI-generated code reveals significant concerns about the long-term maintenance costs associated with using AI coding assistants in software development. Researchers analyzed 302.6k AI-authored commits from 6,299 GitHub repositories, assessing the technical debt introduced by these tools. The study found that a staggering 484,366 distinct issues emerged as a result, with code smells making up 89.3% of the total problems identified. The results alert the AI/ML community to the reality that over 15% of commits from each AI tool included at least one issue, revealing inconsistencies in code quality across different AI assistants. The findings highlight a pressing need for more robust quality assurance protocols in AI-assisted development. With 22.7% of detected issues remaining unresolved in the latest versions of repositories studied, the research suggests that AI-generated code could lead to accumulating technical debt, undermining software reliability and maintainability. This study underscores the importance of monitoring AI contributions to software projects and developing strategies to mitigate potential long-term impacts on code quality, thus guiding future improvements in AI coding tools.