This paper presents a systematic methodology for software development using Large Language Models (LLMs) as execution engines rather than mere assistants. The core contribution is a shift from manual coding to orchestration: the developer acts as an architect and conductor, decomposing systems into natural language instructions, delegating code generation to LLMs, and refining outputs through rapid iterative feedback loops. A second major contribution is context serialization via YAML — a protocol for compressing complex project state into portable, machine-readable blocks that can be transferred between LLM instances, enabling multi-model collaboration and persistent development sessions across context limits. The methodology is demonstrated through a complete case study: building a FastAPI proxy for Groq's LLM service. The demonstration includes code, error handling, iterative refinement, and YAML context transfer. Empirical evidence shows development time reduced by approximately 90% compared to manual methods. This work defines a new discipline: human-led, AI-executed software development. It is not a tutorial but a formal methodology paper, supported by reproducible code and publicly available GitHub repositories.
Khushal Khan Khattak (Wed,) studied this question.