From Stories to Code: Collaborative Modeling and LLMs

LLMs produce poor domain-specific code not because they're incapable, but because they lack domain context. Using Domain Storytelling and EventStorming as structured inputs to LLMs dramatically improves generated code quality. The post walks through three prototype iterations of a recipe-sharing platform (CookWithUs): starting with a Domain Story and Visual Glossary, refining with EventStorming to surface business rules (e.g., 'a Cook cannot rate their own Recipe'), and finally splitting into three bounded-context OpenAPI specs that drive both backend and frontend generation. Each modeling step reduces ambiguity and produces measurably better LLM output. The key insight is that the Ubiquitous Language from collaborative modeling sessions serves as the specification that makes LLM-assisted development reliable — the domain work must still be done by humans, but once encoded precisely, LLMs become effective executors.