Abstract
The evolution of software engineering has increasingly emphasized the optimization and maintainability of enterprise systems, particularly those burdened with monolithic architectures. Traditional refactoring strategies, while effective in improving code readability and reducing technical debt, often fall short when applied to large-scale, complex systems due to limitations in manual effort, expertise constraints, and inherent system interdependencies. Recent advancements in artificial intelligence (AI), particularly through deep learning and large language models (LLMs), have opened novel pathways for automated, intelligent refactoring that not only improves code quality but also enhances overall system resilience and adaptability. This study synthesizes literature on AI-assisted refactoring methodologies, situating these developments within the broader historical and theoretical contexts of software engineering, while critically evaluating their applicability to enterprise monolithic systems. Leveraging Hebbar’s (2023) AI-augmented framework, this article examines the integration of automated pattern detection, semantic code analysis, and predictive modification strategies within industrial-scale software ecosystems. Furthermore, the work highlights the implications of AI-driven interventions for technical debt management, code reuse, and knowledge transfer across development teams. The paper explores methodological considerations, including the design of training datasets derived from open-source repositories, the selection of appropriate neural architectures, and the orchestration of evaluation metrics aligned with software quality attributes. Limitations such as model interpretability, potential overfitting, and contextual sensitivity are discussed, alongside mitigation strategies informed by prior research. Descriptive analysis of case observations illustrates the efficacy of AI-assisted refactoring in reducing maintenance overhead and increasing developer productivity, while also identifying challenges associated with integrating these systems into legacy development pipelines. The discussion extends to future directions, including reinforcement learning for adaptive code modification, the incorporation of explainable AI for regulatory compliance, and multi-modal modeling approaches that consider both code and documentation. This study contributes to the growing discourse on intelligent software engineering by providing a comprehensive, theoretically grounded framework for understanding, deploying, and assessing AI-driven refactoring practices in enterprise environments.