The document discusses different approaches for using SQL in streaming data applications, including structured statements, dbt-style projects, notebooks, and managed runtimes. It evaluates each approach based on criteria like version control, code organization, testability, CI/CD, and observability. Overall, it recommends that for long-running streaming apps, developers should pay special attention to state management, avoid mutability, prioritize integration testing over unit testing, and embrace an SRE mentality. The document also notes that while notebooks are great for exploration, production code is better served by traditional programming frameworks, and that any managed runtime requires excellent developer experience.