2025

Uusnäkki, Joonas; Ihantola, Petri; Mikkonen, Tommi

One potential target area for generative AI is software maintenance. With software that has been maintained for an extended period of time, there is a lot of data available about the design decisions that were made earlier on. At the same time, the maintenance team is usually smaller than the original development team, and certain parts of the system that are maintained may be unfamiliar to the developers involved in the process. Prompt engineering, or crafting input prompts to guide AI models, influences the relevance and precision of the AI’s response and is thus pivotal for applying LLMs in real life. This paper introduces a framework called PESD for creating prompts in maintenance-related tasks. This framework is then used to probe the usefulness of LLMs in an industrial context, using a system that is under active, extended maintenance as the case study.