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