I agree, I bet models could excel at CLI tasks since the feedback would be immediate and in a language they can readily consume. It's probably much easier for t

I agree, I bet models could excel at CLI tasks since the feedback would be immediate and in a language they can readily consume. It's probably much easier for them to to handle "command requires 2 arguments and only 1 was provided" than to do image-to-text on an error modal and apply context to figure out what went wrong.