To What Extent Does Agent-Generated Code Require Maintenance? An Empirical Study (arxiv.org)

A recent empirical study has explored the maintenance needs of code generated by large language model (LLM)-based autonomous coding agents, shedding light on their long-term viability in software development. By analyzing over 1,000 AI-generated files from the AIDev dataset against more than 3,200 changes in popular GitHub repositories, the research reveals that AI-generated code typically requires less frequent maintenance compared to human-authored code. Notably, when maintenance does occur, it primarily involves feature extensions rather than the bug fixes more commonly seen in human-generated code. This study is significant for the AI/ML community as it raises important questions about the sustainability of AI-generated software in real-world applications. With human developers still responsible for the majority of maintenance tasks, the findings suggest that reliance on LLMs for coding may necessitate ongoing human oversight to ensure code quality and reliability. The research encourages further investigation into how AI tools can be improved to reduce maintenance burdens and enhance their integration into conventional software development practices.