Vibe coding—where developers describe intent in natural language and AI generates the code—has gained traction, but its applicability to structured data engineering frameworks like dbt remains underexplored. This experiment tackles a data vault warehouse project using dbt and SQL Server entirely through vibe coding. The author documents the workflow, from defining business concepts to generating dbt models, and evaluates the quality and correctness of AI-produced code. Key findings include that AI handles boilerplate and repetitive patterns well, but struggles with nuanced business logic and dbt-specific conventions like incremental materializations and test configurations. The post offers practical advice on when to trust AI output and when manual review is essential. For data teams considering AI-assisted development, this provides a realistic benchmark of current capabilities and a roadmap for integrating vibe coding into dbt workflows.
An experiment using AI-assisted vibe coding to construct a data vault warehouse with dbt and SQL Server, revealing practical insights for data engineers.