Stop when repeated corrections are not improving the result, the tool lacks the information or capability needed, or checking its work costs more effort than a suitable alternative. Choosing another method is a sensible decision, not a failure to adapt.
Set a small limit before experimenting, such as one initial attempt and one targeted revision. Then compare the outcome with your usual approach. Keep what you learned about the task and the tool. Good AI judgment includes knowing where ordinary software, a reliable reference, or another person is the better resource.