One of the coolest things AI gave me was the ability to build (and constantly review) test scripts that humans can follow. It's funny how much of "let's comprehensively test feature X with Amethyst against client Y" there is now in my day-to-day. AI builds a shell script that walks me through both UIs on Amethyst and on other clients so that I don't lose time learning how other clients work and don't miss branches of code without tests (like multiple settings that affect the feature). The "comprehensive" word is the key. The script controls all test cases, in a combination of all settings, that need to pass without the human having to keep track of which special cases were tested. And since this is Nostr, the script can just watch for an event from either client to verify the results. The human just clicks around.