Agentic Software Issue Resolution with Large Language Models: A Survey
Zhonghao Jiang, David Lo, Zhongxin Liu
TL;DR
The paper addresses the challenge of fully automating software issue resolution with large language models by surveying LLM-based agentic systems that perform long-horizon reasoning, planning, and execution in real-world repositories. It introduces a three-dimensional taxonomy of benchmarks, techniques, and empirical studies, and documents a paradigm shift toward reinforcement-learning–driven agentic designs. Key contributions include a detailed methodology for literature collection, a comprehensive mapping of benchmark construction/evolution/evaluation metrics, and a critical analysis of scaffold designs and learning strategies, supported by empirical insights on benchmark integrity and failure modes. The work highlights open research opportunities in multi-perspective evaluation, richer repository representations, test-driven validation, and SE for agentic systems, aiming to guide future development of robust, autonomous software maintenance agents with real-world impact.
Abstract
Software issue resolution aims to address real-world issues in software repositories (e.g., bug fixing and efficiency optimization) based on natural language descriptions provided by users, representing a key aspect of software maintenance. With the rapid development of large language models (LLMs) in reasoning and generative capabilities, LLM-based approaches have made significant progress in automated software issue resolution. However, real-world software issue resolution is inherently complex and requires long-horizon reasoning, iterative exploration, and feedback-driven decision making, which demand agentic capabilities beyond conventional single-step approaches. Recently, LLM-based agentic systems have become mainstream for software issue resolution. Advancements in agentic software issue resolution not only greatly enhance software maintenance efficiency and quality but also provide a realistic environment for validating agentic systems' reasoning, planning, and execution capabilities, bridging artificial intelligence and software engineering. This work presents a systematic survey of 126 recent studies at the forefront of LLM-based agentic software issue resolution research. It outlines the general workflow of the task and establishes a taxonomy across three dimensions: benchmarks, techniques, and empirical studies. Furthermore, it highlights how the emergence of agentic reinforcement learning has brought a paradigm shift in the design and training of agentic systems for software engineering. Finally, it summarizes key challenges and outlines promising directions for future research.