GitHub’s AI-powered coding assistant, Copilot, may be a game-changer for developers, offering not just speed but quality improvements in coding. A recent study by GitHub delved into the effects of Copilot on code quality, examining its influence on aspects like readability, reliability, maintainability, and conciseness.
Involving a total of 202 Python developers—104 working with the aid of Copilot and 98 operating solo—the study tasked participants with a practical project. They were required to develop a web server for restaurant reviews and test its functionality using ten unit tests. Importantly, the analysis of the code was blind, ensuring reviewers did not know whether AI assistance was involved.
The study yielded promising results, showcasing the significant advantages of AI assistance. Developers using Copilot succeeded in passing all unit tests 56% more often than their non-AI-reliant counterparts. This pointed to a marked increase in functionality. Moreover, Copilot’s influence on readability was notable, allowing developers to write 13.6% more lines of code without losing clarity.
Moreover, the evaluation metrics—spanning readability, reliability, maintainability, and conciseness—displayed an average improvement of 3.29%. The most pronounced benefit was in code conciseness, which was up by 4.16%. Copilot’s impact extended to the code approval process, with AI-assisted code seeing a 5% higher approval rate, thus potentially shortening the time required for coding projects to be production-ready.
These findings underline the potential of AI tools like Copilot in not only expediting the coding process but enhancing code quality, thereby pointing to a future where AI can significantly augment human capabilities in software development.
