Designing Human-Centered Code Completion Tools in the Age of Generative AI

Code Completion Tools (CCTs) powered by Artificial Intelligence are being increasingly used by developers to improve their work performance. However, there is little guidance in the literature about how AI-driven CCTs should be designed to meet developers' expectations of how these systems work, i.e., their mental models. Based on the results of co-design workshops conducted with 56 participants, we formulate user-centered guidelines to design and develop the user interface of and user interaction with CCTs. These emphasize customization, adaptivity, and contextual relevance of code suggestions and explanations inside an AI-powered CCT. Following these design guidelines is expected to create CCTs that better align mental models of developers, resulting in more intuitive, trustworthy, and efficient CCTs.