Is AI-written code buggier than human code? (www.repowise.dev)

Recent research challenges the prevailing notion that AI-generated code is buggier than human-written code. By analyzing the Git history of 28 public repositories with over 112,000 commits, researchers found no conclusive evidence to support claims that AI-written code introduces more defects. In fact, the data indicated that commits from human-driven agents were significantly less likely to introduce bugs, with an odds ratio of 0.57, and agent-written lines had a higher survival rate in the codebase compared to human-written lines. This suggests that when AI-assisted coding is coupled with human oversight, it can produce more reliable and durable code than humans working alone. The study utilized a detailed provenance detection system to categorize commits and employed rigorous statistical measures to ensure the results were not skewed by commit size or other factors. The findings emphasize the importance of human involvement in the coding process, highlighting that human-driven AI agents benefit from the scrutiny of reviews, which enhances code quality. Ultimately, the results indicate that AI tools can be effective partners for developers, supporting a workflow where human judgement combined with AI assistance yields robust code outcomes. This research reshapes the conversation around AI in software development, suggesting that rather than being seen as a source of additional bugs, AI can be a valuable complement to human coding practices.