Autonomous Legacy Web Application Upgrades Using a Multi-Agent System

Valtteri Ala-Salmi, Zeeshan Rasheed, Abdul Malik Sami, Zheying Zhang, Kai-Kristian Kemell, Jussi Rasku, Shahbaz Siddeeq, Mika Saari, Pekka Abrahamsson

2025

Abstract

The use of Large Language Models (LLMs) for autonomously generating code has become a topic of interest in emerging technologies. As the technology improves, new possibilities for LLMs use in programming continue to expand such as code refactoring, security enhancements, and legacy application upgrades. Nowadays, a large number of web applications on the internet are outdated, raising challenges related to security and reliability. Many companies continue to use these applications because upgrading to the latest technologies is often a complex and costly task. To this end, we proposed LLM based multi-agent system that autonomously upgrade the legacy web application into latest version. The proposed multi-agent system distributes tasks across multiple phases and updates all files to the latest version. To evaluate the proposed multi-agent system, we utilized Zero-Shot Learning (ZSL) and One-Shot Learning (OSL) prompts, providing the same instructions for both. The evaluation process was conducted by updating a number of view files in the application and counting the amount and type of errors in the resulting files. In more complex tasks, the amount of succeeded requirements was counted. The prompts were run with the proposed system and with the LLM as a standalone. The process was repeated multiple times to take the stochastic nature of LLM’s into account. The result indicates that the proposed system is able to keep context of the updating process across various tasks and multiple agents. The system could return better solutions compared to the base model in some test cases. Based on the evaluation, the system contributes as a working foundation for future model implementations with existing code. The study also shows the capability of LLM to update small outdated files with high precision, even with basic prompts. The code is publicly available on GitHub: https://github.com/alasalm1/ Multi-agent-pipeline.

Paper Citation

in Harvard Style

Ala-Salmi V., Rasheed Z., Sami A., Zhang Z., Kemell K., Rasku J., Siddeeq S., Saari M. and Abrahamsson P. (2025). Autonomous Legacy Web Application Upgrades Using a Multi-Agent System. In Proceedings of the 20th International Conference on Evaluation of Novel Approaches to Software Engineering - Volume 1: ENASE; ISBN 978-989-758-742-9, SciTePress, pages 185-196. DOI: 10.5220/0013397200003928

in Bibtex Style

@conference{enase25,
author={Valtteri Ala-Salmi and Zeeshan Rasheed and Abdul Sami and Zheying Zhang and Kai-Kristian Kemell and Jussi Rasku and Shahbaz Siddeeq and Mika Saari and Pekka Abrahamsson},
title={Autonomous Legacy Web Application Upgrades Using a Multi-Agent System},
booktitle={Proceedings of the 20th International Conference on Evaluation of Novel Approaches to Software Engineering - Volume 1: ENASE},
year={2025},
pages={185-196},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013397200003928},
isbn={978-989-758-742-9},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 20th International Conference on Evaluation of Novel Approaches to Software Engineering - Volume 1: ENASE
TI - Autonomous Legacy Web Application Upgrades Using a Multi-Agent System
SN - 978-989-758-742-9
AU - Ala-Salmi V.
AU - Rasheed Z.
AU - Sami A.
AU - Zhang Z.
AU - Kemell K.
AU - Rasku J.
AU - Siddeeq S.
AU - Saari M.
AU - Abrahamsson P.
PY - 2025
SP - 185
EP - 196
DO - 10.5220/0013397200003928
PB - SciTePress