From Developer Assistance to Complete Development Automation: A Workflow-Centric Framework for AI Code Generation

Recent advancements in AI-assisted coding tools, such as GitHub Copilot and Cursor, have introduced real-time assistance by suggesting code snippets and improving productivity in isolated coding tasks. However, their scope remains limited to snippet-level assistance, lacking the ability to automate the entire software development lifecycle. This paper proposes a workflow-centric framework for AI-driven code generation that addresses this gap by enabling end-to-end automation—from requirements gathering and code generation to testing, refactoring, and deployment. The framework introduces key components, including feeding the necessary context to LLMs, automatic parsing of generated code, and automatic rollbacks to ensure reliability and scalability. By adopting these foundational building blocks, developers can automate multi-step workflows and streamline repetitive tasks. The reference implementation, JAIG (Java AI-powered Generator), serves as a proof of concept, showcasing how the framework can be implemented. JAIG demonstrates the potential for AI to reshape software engineering practices by moving beyond snippet suggestions toward comprehensive development automation of repetitive tasks.