I agree. These are classic problems where LLMs really shine. I would be very surprised if GPT-5.6 couldn't fix them.
From the minimal context I got, it seems like there are underlying platform access problems in the way. In my experience, attempting to work around these issues is demoralizing.
On the contrary, I think those are fundamental problems that cannot be solved by LLMs.