Analyzing The Use Of Large Language Models In eXtreme Programming Agile Practices

In this dissertation we investigate the utilization of Large Language Models (LLMs) to enhance the effectiveness of Extreme Programming (XP) practices in agile software development. The study delves into various XP practices through a mixed-methods approach combining quantitative and qualitative analysis. The overarching aim is to discern how LLMs can augment human developers’ capabilities within agile practices, examining aspects such as efficiency, collaboration dynamics, code quality, and overall productivity. Our study reveals that the practices that tend to work better with LLMs include those that involve repetitive tasks and require extensive code generation or manipulation, such as Test-Driven Development (TDD) and collaborative programming scenarios. Despite the need for human validation, LLMs can significantly enhance productivity in these contexts. Conversely, practices relying heavily on human judgment, such as user story and use case evaluation, may not benefit as much from LLM integration. We conclude with recommendations for future research such as improving LLMs prompts, and addressing security issues.