AI som kodpartner : En studie om hur mjukvaruutvecklare upplever och använder AI-baserade kodassistenter på arbetsplatsen

Sammanfattning

The development of Artificial Intelligence (AI) has accelerated in recent years andhas impacted multiple industries, including software development. AI code assistantssuch as GitHub Copilot and ChatGPT have become an integral part of softwaredevelopers' workflows and are used for a variety of tasks, including code generation,debugging, and handling repetitive tasks. Previous research has primarily examined how AI code assistants influenceproductivity, code quality, and code security in development processes. Despite theseinsights, there is still a lack of deeper understanding of how software developersthemselves experience and relate to these tools in practice. Therefore, this studyfocused on exploring the factors that influence developers’ experiences and attitudestoward AI-based code assistants. The study is based on a qualitative data collectionmethod, where interviews were conducted with six professional software developersat Visma SPCS to gain deeper insight into their experiences, expectations, andpotential concerns related to the use of AI-based code assistants. Four central themes were identified through thematic analysis of the empiricalmaterial: usage, effect and trust, company culture, and future impact. The results showthat developers use one or more AI code assistants, mainly GitHub Copilot andChatGPT, for code generation, structuring tests, refactoring, and debugging. Despitethe tools’ positive effects, such as increased productivity, several limitations werenoted. Among the most common limitations are AI’s limited contextualunderstanding, lack of support for certain frameworks, hallucinations, and reducedcode quality. The study also shows that company culture significantly influences howand to what extent developers use these tools. Finally, the findings reflect a mixedoutlook on the future, with respondents expressing both optimism and uncertaintyregarding AI’s long-term impact on the software development profession.