This is why it's important that LLMs are implemented as a convenience feature rather than a flow-requirement.

The user needs to be able to see what choices are being made via an LLM (e.g. categorization) and fix those choices when they are wrong.

I'm the OP on X, so LMK if you have any questions or want any context!

This seems even more egregious. You can't even submit a support ticket if the LLM does not deem it worthy.