Actually it would be fun to try to test the developer personality of the models.

Actually there is a nice body of work by Steven Clarke on cognitive dimensions of notations/APIs and the interaction with developer personalities.

I wonder if the same holds for AI models and harnesses.