LLM Agents for Program Repair and Project Setup

Michael Pradel, Stuttgart University

Abstract

Large language models (LLMs) are revolutionizing many aspects of software engineering, by providing rich code understanding, code generation, and code editing abilities. Most past work to leverage LLMs in software engineering interacts with the LLM through an a-priori fixed control flow and by providing a fixed set of information to the model. This talk instead presents an agentic approach, where an LLM interacts with a codebase via a set of tools to autonomously perform a specific software engineering task. We present two such approaches: RepairAgent, which addresses the task of program repair, and ExecutionAgent, which addresses the task of automatically installing and setting up a project so that its tests can be executed. Our results show that agentic approaches outperform prior work, providing a new level of LLM-based automation in software engineering.

Papers

• https://arxiv.org/abs/2403.17134
• https://arxiv.org/abs/2412.10133

Bio

Michael Pradel is 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. The 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, and ASPLOS, and by being named an ACM Distinguished Member.