Structured Prompting for Multi-Stack Project Bootstraps With AI

ShortsMaker is a tool that generates short-form video from saju (Korean astrology) data. Python CLI for processing, React/Remotion for rendering. I bootstrapped the entire project with gpt-5-codex in 4 commits. Starting with code would have failed. Here is the structured approach that worked. Design the Whole Architecture in One Pass When handing a multi-stack project to an AI, the most important rule is: draw the full picture before writing any code. Code-first fails 100% of the time on complex projects. The bootstrap prompt: "Design a ShortsMaker project that takes saju data and generates short-form video. Technical constraints: Python CLI (typer) for data processing pipeline RemixJS for video rendering (remotion) Input: saju data as JSON Output: 9:16 MP4 file Create first: PROJECT_BRIEF.md — project overview and requirements ARCHITECTURE.md — tech stack and component diagram README.md — installation and usage Directory structure and core files Do not: Write all code at once Create c