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

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.