AI tools such as GitHub Copilot can boost the productivity of software developers, but may ultimately put pressure on code quality and the long-term sustainability of software projects. This is the conclusion of new research from Tilburg University. The findings show that AI-assisted programming offers clear benefits, but also introduces new challenges for experienced professionals and the long-term quality of software.
The researchers analyzed how the introduction of GitHub Copilot, a generative AI tool that produces coding suggestions, has reshaped dynamics within open-source software (OSS) projects. Their analysis of 2,755 projects and 1,699 developers shows that AI primarily helps newer, less experienced programmers produce code more quickly. However, this increased productivity comes with a growing maintenance burden: senior developers are forced to review more code, resulting in an average 19% drop in their own productivity. Poonacha Medappa, one of the researchers, explains: “AI tools lower the entry barrier for novice developers, but our research shows that the cost is borne by experienced professionals who are responsible for maintaining software quality. They increasingly shoulder the work needed to safeguard code quality and security. If we’re not careful, we risk creating a system where speed takes precedence over reliability, with all the consequences that entails for the digital infrastructure we all depend on.” The study finds that AI not only increases productivity, but also changes how work is distributed. Experienced developers, the so-called core contributors, spend less time writing new code after the introduction of AI tools and more time reviewing and improving code written by others (or by the AI itself). This results in a 6.5% increase in maintenance and repair tasks, while their own contributions to new code decline. Professor of Operations Management & Technology Jan Fransoo highlights the broader societal implications: “This research is not only relevant for the software industry. It shows that AI, while valuable, can also have unintended effects on how work is organized in other sectors as well, such as healthcare and professional services. If we use AI to boost productivity, we must also invest in the people who safeguard quality. Otherwise, we risk not only lowering the productivity of our most scarce talent over time, but also becoming more vulnerable.” click here to read moreRapid productivity gains, but with risks
Experienced developers shift from innovation to maintenance