Agentic Software Engineering: Leveraging LLMs to Enhance Software Development and Maintenance

CSE Distinguished Seminar

Speaker: Professor Michael Pradel
Faculty member at the CISPA Helmholtz Center for Information Security
and
Full Professor at the University of Stuttgart

Title: Agentic Software Engineering: Leveraging LLMs to Enhance Software Development and Maintenance

Date: Monday, 24 November 2025

Time: 4:00pm - 5:00pm

Venue: Lecture Theater F (Leung Yat Sing Lecture Theater), near lift 25/26, HKUST

Abstract:

Large language models are increasingly being integrated into software engineering tools to assist developers in various tasks. This talk presents our recent work on leveraging LLMs to enhance software development and maintenance. First, we present Testora, the first regression detection approach that uses natural language intent. Given a pull request, Testora generates tests, compares the original and modified behavior, and identifies behavioral changes that are not intended by the developer. Applied to popular Python projects, it detects 19 regression bugs and finds 11 pull requests that unintentionally fix issues, with most confirmed by developers. Second, we introduce ExecutionAgent, an LLM agent designed to automate the setup of complex software projects. Given nothing but the repository URL, ExecutionAgent autonomously installs dependencies, configures the environment, and runs a project's test suite, significantly reducing the manual effort required for project setup. Across 50 projects covering 14 programming languages and diverse toolchains, it succeeds on 33 projects — 6.6x more than prior work — at low cost and without manual intervention. Finally, we study how such agents operate internally through a large-scale analysis of thought-action-result trajectories from leading LLM-based agents. Analyzing 120 trajectories and 2,822 interactions reveals behavioral patterns and anti-patterns that differentiate successful and failed runs, informing more robust agent design, failure diagnosis, and anti-pattern detection. Together, these results demonstrate how LLMs are transforming the software engineering landscape, enabling more intelligent, autonomous, and efficient development and maintenance processes.

Biography:

Michael Pradel is a faculty member at the CISPA Helmholtz Center for Information Security and a full professor at the University of Stuttgart, which he joined after a PhD at ETH Zurich, a post-doc at UC Berkeley, an assistant professorship at TU Darmstadt. He has visited Facebook, UC Berkeley, and UCLA for sabbaticals. His research interests span software engineering, programming languages, security, and machine learning, with a focus on tools and techniques for building reliable, efficient, and secure software. In particular, he is interested in neural-symbolic software analysis, analyzing web applications, dynamic analysis, and test generation. Michael has been recognized through the Ernst-Denert Software Engineering Award, an Emmy Noether grant by the German Research Foundation (DFG), two ERC grants, best/distinguished paper awards at FSE (3x), ISSTA, ASE, ASPLOS, and MSR, and by being named an ACM Distinguished Member.