>Strongly-type datasets do 90% of that, plus they give you Intellisense on the columns in the datatables.

I'm frankly puzzled about this one. As much as I love (and would love) to have intellisense on everything that resides in the database, to "strongly type" datasets assumes that you duplicate your datadictionary in your code no? Doesn't that increase coupling between two components (compiled code and data in the database) that we have worked very hard to de-couple?