Evaluation of AI-Assisted Optimization : The Impact of GitHub Copilot on Web Performance and Usability in React Applications

Sammanfattning

Slow performance in JavaScript-heavy single-page applications is a well-known problem in real-world environments, and rendering inefficiencies are one of the main reasons for poor Core Web Vitals scores. GitHub Copilot is advertised as a tool that can fix it. Previous studies have looked at Copilot for code-completion tasks, but none have measured its effect on frontend architectural restructuring using both performance metrics and user experience data. This study tests whether GitHub Copilot can speed up a slow, poorly built React e-commerce app. The goal is to measure the AI's impact on machine performance, the end-user experience, and the practical challenges developers face when using AI to fix React architecture. Using a mixed-methods approach, this study combines objective Google Lighthouse measurements under strict hardware throttling to reveal real performance bottlenecks, with a usability survey of 140 participants to test whether the speed improvements actually helped users. It records all AI prompts and generated code to document the developer's workflow and any problems faced. Copilot proved to be a highly effective optimizer for specific parts of the application, eliminating main-thread freezing, and an unexpected improvement in accessibility scores was also observed. However, it struggles with global architecture. It caused functional errors, such as deleting important buttons, breaking pagination, and losing file context, but successfully improved the end-user experience. Correlation analysis indicated that the observed Lighthouse performance improvements were associated with higher user satisfaction scores, with the median increasing from 2.80 to 4.27. GitHub Copilot is a fast coding assistant, but it cannot act as an independent software architect. It speeds up the optimization process but forces developers to constantly check for AI mistakes. It can significantly improve performance and usability in React e-commerce applications with similar characteristics, but only if a skilled human developer supervises the process.