Advancing Software Engineering Through Generative AI and Retrieval-Augmented Generation for Automated Code Development and Intelligent Debugging

Abstract

Generative artificial intelligence is changing each stage of the software development lifecycle, from understanding requirements and writing code to generating tests, resolving issues, and maintaining systems in production. Most research and tooling, however, concentrates on isolated tasks such as completing a line of code or solving a self-contained programming problem. Enterprise software development is a lifecycle activity carried out on large, interconnected codebases, and success depends on knowledge that general-purpose models do not have. This article takes a lifecycle perspective on generative AI in software engineering. It traces the progression of AI for code from learned code representations to competition-level generation. It explains why retrieval-augmented generation (RAG) is essential for applying these models to real projects. It then proposes an AI-augmented lifecycle model that assigns each development stage a defined AI role, a retrieval source, and a human checkpoint. The article also introduces a three-level adoption model, from assistive to supervised autonomous use, and illustrates both models with a feature development case in an enterprise asset-monitoring system. It concludes that organizations gain the most from generative AI when they treat it as a lifecycle capability, grounded in project knowledge and matched to appropriate levels of human oversight.