Artificial Intelligence-Assisted Refactoring and Architectural Transformation: A Comprehensive Study on Large Language Models for Migrating Monolithic Software Systems to Microservices

Abstract

The rapid evolution of software systems and the increasing complexity of enterprise-scale applications have intensified the need for architectural modernization. Traditional monolithic software architectures, while historically dominant due to their simplicity and unified deployment model, have become increasingly difficult to maintain, scale, and evolve in contemporary distributed computing environments. Consequently, microservices architecture has emerged as a widely adopted paradigm for building scalable, modular, and independently deployable systems. However, the migration from monolithic architectures to microservices presents significant technical and organizational challenges, including service decomposition, dependency identification, data migration, and architectural restructuring. In recent years, the emergence of large language models (LLMs) and AI-assisted development tools has introduced new possibilities for automating aspects of software engineering, including code generation, code analysis, documentation, and architectural refactoring. This research article investigates the role of LLMs in facilitating the transformation of monolithic systems into microservices-based architectures. Drawing upon a comprehensive set of scholarly references on large language models, AI-assisted programming, prompt engineering, software refactoring strategies, and microservices extraction techniques, the study provides a detailed theoretical and methodological exploration of AI-augmented architectural transformation. The article examines how LLM-driven tools can assist developers in identifying service boundaries, summarizing code structures, generating refactoring suggestions, and supporting iterative system decomposition. Additionally, the study analyzes the challenges associated with AI-assisted development, including hallucination, non-determinism in code generation, prompt sensitivity, and security vulnerabilities such as prompt-based exploitation. Through a conceptual methodological framework and interpretive analysis of existing research findings, the article evaluates both the capabilities and limitations of LLM-based software engineering assistance. The discussion further explores implications for future research, particularly regarding the integration of AI agents into automated software modernization pipelines. Ultimately, the article argues that while LLMs do not fully replace traditional software engineering practices, they represent a transformative tool that can significantly enhance developer productivity and accelerate architectural modernization when used responsibly within structured engineering workflows.