Honestly the only thing I see current level AI reliably doing without being used as an assistive tool is grunt info work. For example a lot of T1 customer service positions can moreorless be replaced out with the current level of LLM's that we have. Many T1 support roles consist almost entirely of searching the current customers issue, copy/pasting a boilerplate solution list of what may fix it, asking "did that work?" and if not escalating to the next tier. Hallucinations at this level won't have a very big impact outside of annoying the customer and the t2 when it gets escalated because it failed to fix the issue. Said system shouldn't have control over anything, it should strictly be information based. Anything management wise or financial wise or general output of merchandise should not be using these technologies standalone, at most it should be an assistive tool to a human in that position.