Yes. I already use GPT-4 to replace support for a 3D CAD system I use, since it is usually takes a significant time & effort to get through to tech support, and this often provides very good answers (up to 2021).
I think the best solution would be for them to train and implement a version of GPT-4 or similarly capable LLM on all of their documentation and support logs, then keep the live users to handle the issues where the LLM fails and to continue training it on new info.
(But of course, management will likely just dump the people before it is even time, as with the examples in the article, following the ensittification principle of monetization of evertyhing. Seems like that was missed in the Ferengi Rules of Acquisition.)