AI-assisted data development is transforming SQL writing, utilizing either direct SQL generation or a layered approach with intermediate representations. This article focuses on a combined model of Trae (AI planning) and SQLazy (execution layer). This partnership establishes a "AI planning + human review + deterministic engine execution" framework. The process involves Trae generating step-by-step SQLazy scripts (.nspl) after loading project knowledge, clarifying requirements, and decomposing tasks. SQLazy's IDE facilitates step-by-step execution, real-time debugging, and cross-database compilation from a single script.The workflow begins with environment setup, followed by triggering Trae with detailed business requirements. The most crucial phase is validation and correction in the SQLazy IDE, where test data is used to verify intermediate results. Four real-world case studies illustrate the methodology: statistical aggregation, multi-table merging, cross-subgroup data filling, and amount allocation. While simpler cases were correctly handled in one go, complex scenarios like cross-group data filling and amount allocation required iterative corrections. The cross-group filling case highlighted the importance of using row numbers as a mapping bridge when direct join keys are unsuitable. The amount allocation case, the most complex, required multiple iterations to refine syntax and logic, particularly concerning rounding and remainder distribution to ensure total preservation.