The problem with the use of AI-like tools (LLMs like ChatGPT aren't really AI, they're an AI simulation - more like "predictive text on steroids", really) isn't their existence, but rather being put to inappropriate uses.
When they're used to save time by people who are already skilled enough to do the job in the first place, or supervised heavily by humans, they're brilliant.
The problem comes when they're used to replace humans in an interaction chain...and that seems to be their purpose as interpreted by management just about everywhere.
In the context of the article, it strikes me that the main problem is when it misses critical information - which it absolutely will do. Unlike a deterministic system, where there would be an audit trail allowing somebody to fix the bug that led to the omission, the "AI" in question will simply be a black box. They can try to influence its behaviour, but there are absolutely no guarantees about how it will behave in future.
It's a terrible idea, but one which will be forced into existence everywhere by people who have no business making those decisions due to a comprehensive lack of knowledge and wilful ignorance of facts in favour of marketing potential.