There are a lot of elements of Beta that AI can test MUCH better than I can.
List night, I set up a test for one of the current game systems–the method by which you level up letter tiles.
See, when you answer questions in Beta and defeat mobs, you get more letter tiles. They’re a kind of loot. They all start of as basic, level 1 tiles, but you can merge them to make them stronger. 2 level 1 “A” tiles make 1 level 2 “A” tile. 2 level 2 “A” tiles make a level 3 “A” tile. Etc.
(I don’t, as of yet, have an upper bound for that mechanic. Is it necessary? Maybe not. It’s exponential growth, right? That’s a very natural boundary.)
Getting a level 11 tile for ANY letter would require 2048 basic tiles of that letter. I don’t know if anyone’s going to do that. I don’t even know if it’s POSSIBLE.
So, I told Claude to run the game and play it as a variety of gamers who spend a variety of time playing the game. I wanted him to go for a month (or longer?) with each “test player” min/maxing their build to prioritize getting their letter tiles leveled up as high as possible.
Just to see how far it would go. Is there a natural cap, and how long does it take to reach it?
So, I set Claude to work. Thinking this test would take a while (Claude guessed a few hours, but his guesses are wildly inaccurate), I set it to run right before bed. Had Claude ask clarifying questions before it started. Had a good, detailed test plan. I hit Enter and off it went.
I wake up in the morning to see something like this:
Claude needed permission to delete files from the git repository, popped up a window asking for permission, and then just sat there for over 6 hours doing nothing.
So, the test didn’t run.
Considering the number of times that repository gets updated, there really should be a means of allowing Claude to work with git and not have to ask permission each time.
Frustration.