Large language models (LLMs) are radically changing everyday life, and anybody can leverage them for various tasks. Such tasks also include software development-related ones. This has changed how software engineers conceive, implement, and evolve their software. Also, this has profoundly impacted software engineering research, making some of the previous achievements obsolete and forcing everybody to rethink what would have been valuable contributions in this context. What about software engineering education? In this talk, I will outline my experience in embedding AI-intensive tools in software engineering courses, highlighting the positive aspects, challenges and risks I learned from the feedback I received from my students.

Massimiliano Di Penta is a full professor at the University of Sannio, Italy. His research interests include software maintenance and evolution, software analytics, empirical software engineering, and artificial intelligence for software engineering. He is an author of over 320 papers that appeared in international journals, conferences, and workshops. His h-index is 67 (Scopus), and 91 (Google Scholar). He has received several awards for his research and service, including a SIGSOFT Distinguished Service Award (2004), six ACM SIGSOFT Distinguished paper awards, and five most influential paper awards. According to articles that appeared in the Journal of Systems and Software, he was considered the second-most productive expert researcher in the software engineering area for the period 2010-2017 and for the period 2017-2020. He serves and has served in the organizing and program committees of more than 120 conferences, including the three most important conferences in the research area of software engineering (ICSE, ESEC/FSE, and ASE). He has been program co-chair of ICSE 2023, ESEC/FSE 2021, ASE 2017 and other software engineering conferences. He is associate editor-in-chief of IEEE Transactions on Software Engineering, co-editor in chief of the Journal of Software: Evolution and Processes edited by Wiley, editorial board member of Empirical Software Engineering Journal edited by Springer. He has served the editorial board of the ACM Transactions on Software Engineering and Methodology and of the IEEE Transactions on Software Engineering.

Gabriele Keller is the head of the software engineering division and the chair of the software technology group at Utrecht University. Her research interest include programming language technology, in particular for functional and embedded languages, software verification, as well as high-performance computing. She received her PhD degree in 1999 from the Technical University, Berlin, on the implementation of irregular data-parallelism. She worked at the University of New South Wales in Australia, where she co-founded the programming language group, and moved to Utrecht University in 2018.

Model-based testing has been the promising next step in test-automation for a while now; and while it has been picked up as a software engineering practice, it is not applied at a large scale. On the other hand, Behaviour-Driven Development has become popular recently in software development. As part of this approach, scenarios describing a behaviour of the system are written in a structured natural language format. In my talk, I will present an approach where we translate these scenarios into a formal model, such that the test-automation techniques of model-based testing can be exploited.

With applications becoming more complex and higher-level, we need compilers to keep up with these higher levels of complexity and abstractions. This talk focuses on how different levels of abstraction in compiler intermediate representations (IRs) can improve optimization techniques. We’ll discuss how abstractions make it easier to understand and implement optimizations. In particular, how abstraction can turn global, effectful reasoning into local, pure reasoning. We also discuss how optimizing and exploring these large design spaces in compiler development, with a focus on the tradeoffs between automation and manual optimization.

This talk outlines the practical application and assessment of large language models tailored to the code generation task. It will cover our approach for evaluating these models’ performance in real-world software development environments and how we anlayzed the failed cases. I will also present promising research areas to address the existing challenges and improving these models.