Totally agree! If you can find a working xpath locator, then this is better than using a computer vision or OCR. But sometimes finding an xpath is tricky, for example if the part you need to monitor is inside an iframe.
Also, sometimes I have some “quick and dirty” automation tasks where the macro runtime or efficiency is not crucial. For these RPA tasks I tend to use computer vision instead because it makes creating the automation much faster 
Lately I even experiment with the new aiPrompt and aiComputerUse commands. Their big downside is that you need to buy anthropic api credits. so that is only good for tasks that need to run only once or twice or it becomes really expensive. But for these one off tasks, it makes creating the macro even faster.