It is an extremely interesting thought.

I bet the level of testing is much lower especially when considering that this area of research is undoubtedly less mature and therefore we will have less input into professional development.

When I have done similar work (i.e. generated code) I have not been as rigorous as I otherwise would have been as I have been less confident in ways to provide the same level of coverage.

I, for one, am certainly interested in any experience (successful or less so) in this area. Also not necessarily limited to SQL related features (SSIS, ETL etc.) as often cross pollination of ideas works wonders.