Large Language Models (LLMs) have shown significant promise in various software engineering (SE) tasks. However, their integration into broader software engineering processes presents unique challenges, particularly due to their limited understanding of domain-specific knowledge and the structured nature of SE data. In this talk, I will highlight key lessons learned and challenges encountered while applying LLMs to software engineering. This includes the necessity of filtering out noisy data and the benefits of combining LLMs with program analysis techniques to enhance their grasp of context and structure. Additionally, I will explore the transformative effects of LLMs on various software engineering practices, including test case generation, vulnerability management, and automated code generation. The presentation will provide an in-depth look at both the limitations and possibilities of LLMs in software engineering, offering insights into emerging opportunities and future trends in the field.

Xin Xia is a Qiushi Distinguished Professor at the College of Computer Science and Technology, Zhejiang University. Prior to joining Zhejiang University, he was the Chief Expert and Director of the Software Engineering Application Technology Lab at Huawei Technologies, China. His research interests encompass AI and SE, mining software repositories, and empirical software engineering. Xin has authored over 360 papers and has received 16 best or distinguished paper awards, including nine ACM SIGSOFT Distinguished Paper Awards for his contributions at ASE (2018-2021), ICPC (2018, 2020), FSE (2025), ICSE (2024), and MSR (2024). In recognition of his early career achievements, he was awarded the ACM SIGSOFT Early Career Researcher Award in 2022. Xin is an ACM Distinguished Member for contributions to empirical software engineering and AI for software engineering.