LLMs in Software Design: GitHub Study and Practitioner Survey
An empirical study has been conducted to explore the utilization of Large Language Models (LLMs) by software developers in the realm of software design. This research integrates an analysis of 291 conversations between developers and ChatGPT on GitHub with a survey involving 65 professionals. It uncovers nine distinct categories of design tasks that ChatGPT assists with, such as architecture design, data model design, and design patterns. The findings depict the nature of interactions between developers and ChatGPT, highlighting that developers mainly leverage ChatGPT for targeted design tasks. The objective of the paper is to enhance the understanding of LLM applications in software design, a field that has typically depended on human knowledge.
Key facts
- Study mined 291 developer-ChatGPT conversations on GitHub
- Surveyed 65 software practitioners
- Identified nine categories of design tasks supported by ChatGPT
- Tasks include architecture design, data model design, and design patterns
- Research focuses on LLM usage in software design
- Software design traditionally depends on human expertise
- Paper published on arXiv with ID 2605.01392
- Study uses mixed-methods approach
Entities
Institutions
- GitHub
- arXiv