# Agents in Software Engineering: Survey, Landscape, and Vision

**Authors:** Yanlin Wang, Wanjun Zhong, Yanxian Huang, Ensheng Shi, Min Yang, Jiachi Chen, Hui Li, Yuchi Ma, Qianxiang Wang, Zibin Zheng  
**arXiv:** [2409.09030](https://arxiv.org/abs/2409.09030)  
**Date:** 2024-09-13  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2409.09030v2/

## Key points

- The paper explores the intersection of Large Language Models (LLMs) and software engineering (SE), emphasizing the growing use of LLMs in various SE tasks.
- It highlights a lack of detailed analysis on how LLMs are combined with SE to optimize tasks and the framework guiding their integration.
- The authors conducted a survey to examine existing works that combine LLM-based agent technologies with SE, introducing a framework comprising perception, memory, and action modules for LLM-based agents in SE.
- They discuss current challenges in merging LLMs with SE and propose future opportunities to overcome these obstacles.
- The authors maintain a GitHub repository of related papers to facilitate further research and collaboration in this field.

## Summary

The paper delves into the intersection of Large Language Models (LLMs) and software engineering (SE), highlighting the increasing use of LLMs in various SE tasks. The authors point out that while many studies have integrated LLMs with SE through the concept of agents, there is a lack of detailed analysis on how these technologies are combined to optimize tasks and the framework guiding their integration. To address this gap, the authors conduct a thorough survey to examine existing works that combine LLM-based agent technologies with SE. They introduce a framework for LLM-based agents in SE comprising three key modules: perception, memory, and action. This framework serves as a guide for understanding how LLMs can be effectively utilized within an agent-based system in the context of SE tasks. Additionally, they discuss current challenges in merging LLMs with SE and propose future opportunities to overcome these obstacles. By maintaining a GitHub repository of related papers on this topic at https://github.com/DeepSoftwareAnalytics/Awesome-Agent4SE., they aim to facilitate further research and collaboration in this evolving field. Overall, "Agents in Software Engineering: Survey, Landscape, and Vision" provides valuable insights into the integration of LLM-based agents with SE tasks and offers a roadmap for future developments in this area.