Limitations and Technological Risks of Using Language Models for Refactoring Legacy Software Systems

Citation: Sergei Kuznetsov, "Limitations and Technological Risks of Using Language Models for Refactoring Legacy Software Systems", Universal Library of Engineering Technology, Volume 03, Issue 02.

Copyright: This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Abstract

This article provides a comprehensive analysis of the technological limitations and risks associated with the application of large language models in the refactoring of legacy software systems. The study is conducted as a structured review and analytical synthesis of peer-reviewed publications devoted to AI-assisted software engineering, automated refactoring, and the modernization of legacy codebases. The analysis focuses on empirical and conceptual studies that examine the capabilities of language models in code transformation tasks and their interaction with complex software architectures. Particular attention is paid to the influence of contextual constraints, architectural dependencies, and accumulated business logic typical for legacy systems, which significantly complicate automated code transformation. The findings show that while large language models can effectively improve certain structural characteristics of code and assist in eliminating repetitive design defects, their performance becomes unstable when transformations affect functional semantics and inter-module dependencies. It is established that improvements in code quality metrics do not necessarily correspond to the preservation of program behavior, which creates additional technological risks during modernization. The paper proposes a structured classification of risks associated with LLM-based refactoring and demonstrates that the most reliable approach currently lies in hybrid workflows combining automated code suggestions with human validation, testing, and static analysis. The results contribute to a better understanding of the safe integration of intelligent assistants into software modernization processes and may inform the design of risk-aware AI-supported refactoring practices.

Keywords: Large Language Models; Software Engineering; Code Refactoring; Legacy Software Systems; Software Modernization; Artificial Intelligence.

https://doi.org/10.70315/uloap.ulete.2026.0302004