What criteria do you use to decide which test cases should be automated and which should remain manual?
