Agentic Software Issue Resolution with Large Language Models: A Survey

2025-12-24

AI Summary

This paper addresses real-world software engineering challenges—such as bug fixing and performance optimization—that demand long-horizon reasoning, iterative exploration, and feedback-driven decision-making by LLM-based agents. Methodologically, it synthesizes 126 recent studies, integrating ReAct and Plan-and-Execute architectures, multi-step tool invocation, interactive code-environment execution, and benchmarks including SWE-bench. Its primary contribution is the first domain-specific, three-dimensional taxonomy of agent capabilities for software engineering—spanning benchmarks, techniques, and empirical evaluation—which reveals a paradigm shift toward reinforcement learning–driven agent design. The study maps the field’s evolutionary trajectory, identifies critical bottlenecks—including sparse environmental feedback and low planning interpretability—and proposes six concrete future directions: scalable training, simulation-augmented learning, human-in-the-loop collaboration, modular agent design, benchmark standardization, and causal reasoning integration.

Technology Category

• Cognitive Modeling & Cognitive Systems: Agent Architectures
• Search and Optimization: Metareasoning and Metaheuristics
• Multiagent Systems: Agent/AI Theories and Architectures

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Agentic search
• Economics, Online Markets and Human Computation: Architectures and workflows that use LLMs for crowd work

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.

Problem

• Surveying LLM-based agentic systems for software issue resolution
• Addressing complex software maintenance with long-horizon reasoning
• Enhancing software quality via agentic reinforcement learning paradigms

Innovation

• LLM-based agentic systems for software issue resolution
• Agentic reinforcement learning for paradigm shift
• Systematic survey of 126 studies on agentic techniques