Generative AI for Software Engineering: From Research Evidence to Industrial Impact

Dario Di Nucci
Associate Professor, University of Salerno

Dario Di Nucci is an associate professor in the Department of Computer Science at the University of Salerno. His research focuses on empirical software engineering, particularly on software testing of functional and non-functional requirements, including security and sustainability, where he applies artificial intelligence, quantum computing, and data mining. He serves as an associate editor for ACM Transactions on Software Engineering and Methodology, Elsevier Journal of Systems and Software, and IEEE Software. He is also a program committee member of international conferences and a referee for international journals in software engineering and artificial intelligence. For these services, he has received several awards. He is a member at large of the IEEE Computer Society Conference Activities Committee and a member of the promotion commission of the Embedded Systems and Smart Manufacturing Lab of the Consorzio Interuniversitario Nazionale per l’Informatica.

Personal Page

Demystifying LLMs for Smart Contract Security: How Model Architecture and Prompt Design Shape Vulnerability Detection

Fabio Calefato
Associate Professor, University of Bari

Fabio Calefato is an Associate Professor at the Department of Computer Science, University of Bari, Italy, and co-founder & CEO of PeoplewareAI, an academic spin-off transferring AI and software engineering research to industry. His research focuses on collaborative and social software engineering, mining software repositories, SE4AI, and, more recently, the adoption of generative AI in software development practice and research. He serves on the editorial boards of EMSE, JSS, and ASE, and has held organizing roles at major SE conferences, including Program Co-chair of ICSME 2024 and CHASE 2023.

Personal Page

From Hype to Evidence: How GenAI is Reshaping Code, Developers, and Research

Generative AI is surrounded by bold claims of collapsing code quality, effortless productivity, and research done by machines. But does close-up evidence match the narrative? This talk examines GenAI in software engineering “in the wild” through three lenses: what it does to real codebases, how developers adopt it and how it changes them, and how researchers both study and use it. Drawing on five recent studies, the talk shows where measured reality contradicts the headlines: code churn does not surge after adoption, sustained use is driven by individual rather than organizational factors, and AI-assisted coding is a cognitively distinct activity. The closing message is that rigorous measurement and reporting are what turn GenAI results into evidence.

Fabio Palomba
Associate Professor, University of Salerno

Fabio Palomba is an Associate Professor at the Software Engineering (SeSa) Lab of the University of Salerno. He received the European PhD degree in Management & Information Technology in 2017. His PhD Thesis was the recipient of the 2017 IEEE Computer Society Best PhD Thesis Award. His research interests include software engineering with and for artificial intelligence, software maintenance and evolution, and empirical software engineering. He has received multiple distinguished and best paper awards at top-tier software engineering venues, serves on the editorial boards of leading journals in the field, and was awarded the 2023 IEEE/TCSE Rising Star Award. In 2019 he was the recipient of an SNSF Ambizione grant, one of the most prestigious individual research grants in Europe.

Personal Page

Prompts as a Language: Toward the Science of Prompt Engineering

Communication has always been one of the defining forces behind human evolution. Our ability to exchange knowledge, negotiate intentions, and coordinate complex activities has enabled the creation of increasingly sophisticated technologies, including software itself. Today, however, software development is undergoing a profound transformation. Large Language Models and autonomous AI agents are rapidly becoming active collaborators in the engineering process, assisting developers in tasks ranging from requirements elicitation to code generation, testing, and maintenance. As AI systems take on greater responsibilities, software engineers are transitioning from being the sole producers of software to becoming orchestrators of increasingly autonomous collaborators.

This shift raises a fundamental question: How should we communicate with these collaborators?

Current prompt engineering practices are largely based on empirical heuristics and trial-and-error, despite growing evidence that principles from human communication, such as Theory of Mind, Socratic reasoning, and cognitive framing, can significantly influence AI behavior. This talk argues that these insights, while valuable, represent only the beginning. AI agents do not ultimately “understand” language as humans do; they operate through complex neural representations whose behavior emerges from statistical learning rather than human cognition. Consequently, prompt engineering requires a stronger scientific foundation. I will introduce the vision of Intent Engineering, a research direction that aims to engineer, represent, and verify developer intent to improve the trustworthiness of AI-assisted software engineering. Building on this perspective, I will argue that prompts should be treated as a new form of programming language rather than as unstructured natural language. This opens the door to applying decades of software engineering knowledge, including syntax and semantic analysis, abstract syntax trees, formal methods, and visual languages, to systematically understand, design, and verify interactions with AI agents. The talk concludes by outlining a research agenda toward a science of prompt engineering, where communication with AI becomes an engineering discipline grounded in rigorous principles rather than an art driven by intuition.

Gemma Catolino
Assistant Professor, University of Salerno

Gemma Catolino is an Assistant Professor at the Software Engineering (SeSa) Lab, Department of Computer Science, University of Salerno, Italy. She received her European Ph.D. in Computer Science from the University of Salerno in 2020. Her research focuses on software engineering for AI, social software engineering, software maintenance and evolution, empirical software engineering, source code quality, and mining software repositories, with a particular emphasis on data-driven and machine learning-based approaches. She has received multiple Best Paper Awards at leading international software engineering conferences. She has served in several organizational roles, including General Chair and Program Co-Chair for international conferences such as MOBILESoft and Euromicro, as well as Track Chair for venues including MSR 2025 and ICSME 2026. Since 2022, she has been an Editorial Board Member of Science of Computer Programming and SoftwareX, a Review Board Member of the Journal of Systems and Software, and, since 2026, an Editorial Board Member of IEEE Transactions on Software Engineering. She is also a member of the Board of Distinguished Reviewers of ACM Transactions on Software Engineering and Methodology (TOSEM) and has received multiple Distinguished Reviewer Awards in recognition of her reviewing service.

Personal Page

Beyond Accuracy: Engineering Fairness for Generative AI in Software Engineering

Generative AI is reshaping software engineering, enabling developers to automate coding, testing, documentation, and many other development tasks. However, accuracy and productivity alone are not enough to ensure responsible AI adoption. Fairness has become a key quality attribute, requiring software engineers to address bias throughout the entire AI lifecycle rather than only at the model level. This talk presents recent research on engineering fairness, covering fairness-aware software engineering practices, empirical evidence from practitioners, and the emerging challenges introduced by Generative AI. Building on these findings, the talk discusses why fairness should be treated as a continuous software engineering concern and outlines future research directions for developing Generative AI systems that are not only effective, but also fair, trustworthy, and human-centered.

Leonardo Mariani
Full Professor, University of Milano – Bicocca

Leonardo Mariani is a Full Professor at the University of Milano-Bicocca. He holds a PhD in Computer Science, received from the same university. His research interests include software engineering, in particular software testing, program analysis, automated debugging, AI4SE, and self-adaptive systems. He has authored more than 150 papers appeared at top software engineering conferences and journals. He received the ICST MIP award for his ICST 2013 paper. He was awarded the ERC Consolidator Grant in 2015, an ERC Proof of Concept grant in 2018, and he is currently active in several national and international projects. He is regularly involved in the organization of major software engineering conferences.

Personal Page

Getting the Right System, Getting the System Right, Getting EveryThing: Experiences with project-scale Gen AI

Software development is being reshaped by agentic systems, which aim to accelerate application development through LLM‑based agents capable of assisting developers throughout the development process. While their effectiveness in small, localized generation tasks is well established, automated code generation at full project scale remains a challenging scenario. This talk presents results and insights from a set of experiments on repository‑scale code generation with agentic systems, highlighting both what these systems currently excel at and where they struggle. Particular attention is given to the distinction between obtaining the right system (getting the right system) and obtaining any working system (getting the system right), illustrating how agentic approaches behave with respect to these two objectives.

Luciano Baresi
Full Professor, Politecnico di Milano

Luciano Baresi is Vice Rector for the Cremona Campus and Full Professor at Politecnico di Milano, within the Dipartimento di Elettronica, Informazione e Bioingegneria (DEIB), where he earned both his Laurea (M.Sc.) and Ph.D. in Computer Science. He has held visiting positions at the University of Oregon (USA), Tongji University (China), and the University of Paderborn (Germany).

Prof. Baresi has served as General and Program Chair for several leading international conferences. He currently serves as Editor-in-Chief of the Proceedings of the ACM on Software Engineering, Senior Associate Editor of ACM Transactions on Autonomous and Adaptive Systems, and is a member of the editorial boards of Science of Computer Programming, IEEE Software, the International Journal of Cooperative Information Systems, and Requirements Engineering.

He has co-authored more than 200 scientific publications and one book (in Italian). His research spans a broad range of software engineering topics. Early in his career, he focused on formal modeling and specification languages, later shifting toward UML and web application design. His current research interests include self-adaptive systems, dynamic software architectures, edge computing, and AI/ML-based systems.

Personal Page

Can LLMs Generate User Stories and Assess Their Quality?

Requirements elicitation is still one of the most challenging activities of the requirements engineering process due to the difficulty requirements analysts face in understanding and translating complex needs into concrete requirements. In addition, specifying high-quality requirements is crucial, as it can directly impact the quality of the software to be developed. Although automated tools allow for assessing the syntactic quality of requirements, evaluating semantic metrics (e.g., language clarity, internal consistency) remains a manual and time-consuming activity. This talk reports the results of experiments conducted to understand how LLMs can help automate requirements elicitation within agile frameworks, where requirements are defined as user stories (US). We used 10 state-of-the-art LLMs to investigate their ability to generate US automatically by emulating customer interviews. We evaluated the quality of US generated by LLMs, comparing it with the quality of US generated by humans (domain experts and students). We also explored whether and how LLMs can be used to automatically evaluate the semantic quality of US. Our results indicate that LLMs can generate US similar to humans in terms of coverage and stylistic quality, but exhibit lower diversity and creativity. Although LLM-generated US are generally comparable in quality to those created by humans, they tend to meet the acceptance quality criteria less frequently, regardless of the scale of the LLM model. Finally, LLMs can reliably assess the semantic quality of US when provided with clear evaluation criteria and have the potential to reduce human effort in large-scale assessments.

Riccardo Coppola
Assistant Professor, Politecnico di Torino

Riccardo Coppola received his PhD in 2019 and is Assistant Professor of Software Engineering at Politecnico di Torino. His main research interest are the application of AI to software testing and requirements elicitation, and the evaluation of fairness of AI systems.

Enhancing Requirements Engineering with Large Language Models and Retrieval-Augmented Generation

Requirements Engineering is a key activity for defining software systems, but it is still largely based on natural language artefacts that may introduce ambiguity, incompleteness, inconsistent terminology, and difficulties in validation. Recent advances in GenAI offer new opportunities to support analysts and stakeholders in the elicitation, refinement, and modelling of requirements. However, fully automated requirements engineering remains risky, especially when domain knowledge, stakeholder intent, and terminology consistency must be preserved. This research explores how Large Language Models and Retrieval-Augmented Generation can support human-in-the-loop Requirements Engineering by combining rule-based quality checks, structured prompting strategies, and project-specific knowledge sources. As a first step toward this broader objective, we investigated LLM-based requirements elicitation, goal extraction, and glossary-aware requirement refinement. We present initial results and discuss future directions for using GenAI and RAG to improve requirement quality, consistency, and maintainability.

Tommaso Fulcini
Postdoc researcher, Politecnico di Torino

Dr Tommaso Fulcini received the M.Sc. degree in computer engineering, Software branch, from Politecnico di Torino, Turin, Italy, where he recently obtained the PhD degree. His research interests include gamification in Software Engineering, particularly software testing. Other research interests include agentic AI for software engineering practices, code quality, and test repair. He is part of the organising committee of Gamify 2026, ASTA 2026 and the Self-Driving Car Tool competition 2026, hosted by ICST and SBFT conferences and a PC member of international conferences in software engineering (ISSTA, ASE, ICSE, FSE).

Personal Page

LABBGEN: LLM-based Automated Black-Box Test Generator

In the context of advanced manufacturing, where software quality for production management is critical, LABBGEN is an innovative project aiming at automating the generation of black-box system tests. In many industrial environments, the lack of detailed documentation and limited access to internal system data or source code make the creation of reliable black-box test suites difficult and time-consuming.

LABBGEN seeks to improve the quality and reliability of manufacturing software by reducing the effort required to produce test cases and documentation, while easing the human effort in creating effective test cases and increasing the number of software defects identified during testing.

LABBGEN adopts a multi-agent LLM-based architecture composed of specialised components responsible for planning, executing, and refining test cases. These agents analyse software behaviour, observe interactions between system modules, and generate test cases that simulate realistic user actions and production workflows. The platform allows also to rely on heterogeneous inputs, such as system screenshots, formal documentations, high-level descriptions, to complement the knowledge of the system and generate fully functional test cases.