Integrating Large Language Model–Driven Code Generation and Business Process Automation: Impacts on Enterprise Systems Performance and Maintainability

Abstract

The increasing maturity of large language model (LLM)–based code generation tools offers unprecedented opportunities to accelerate software development, particularly in domains requiring rapid prototyping, business process automation, and enterprise application deployment. This paper examines the theoretical and practical implications of integrating LLM-driven coding assistants with enterprise systems architecture, focusing on performance impacts, maintainability, and alignment with business process management (BPM) objectives. Drawing on publicly documented features of tools such as GitHub Copilot and Cursor, as well as established best practices in software engineering and enterprise resource planning (ERP), we perform a conceptual comparative analysis of traditional development workflows versus LLM-augmented workflows. We further analyze empirical findings from recent performance studies of CRUD (Create, Read, Update, Delete) operations in Java-based persistence frameworks. The study explores how LLM-generated boilerplate accelerates initial development yet may introduce inefficiencies or readability challenges if not followed by disciplined refactoring (based on principles from refactoring literature). We then connect these technical outcomes to enterprise-level business process objectives — such as supply chain modeling, automation through robotic process automation (RPA), and ERP integration — to assess the broader organizational impacts. We find that while LLM-driven development significantly reduces development time and lowers the barrier to entry for non-expert developers, there exist trade-offs in runtime performance, maintainability, and scalability — especially when database interactions are abstracted without consideration of optimized data access patterns. To manage these trade-offs, we propose a hybrid workflow combining LLM-assisted code generation, systematic refactoring, and performance validation as part of the release pipeline. We conclude with a roadmap for future empirical evaluation and integration strategies to maximize business value while minimizing technical debt.