Automating the Software Development Lifecycle: A Multi-Agent Workflow Powered by Medium-Sized Language Models

Abstract

The Auto Coding System is an innovative framework designed to automate the entire software development lifecycle, from requirements gathering to documentation. This research demonstrates that complex software development processes can be effectively automated while maintaining professional-grade output. By leveraging medium-sized language models (24B-32B parameters) and a structured seven component workflow, the system achieves significant reductions in development time (68% average) while maintaining high-quality standards. The system balances computational efficiency and robust language understanding to achieve seamless collaboration among agents. Critical decision points for requirements understanding and code quality create feedback loops that minimize errors and ensure a high standard of deliverables.