I'm not saying this is you, but in my experience, the conversation changes because people flatly refuse to believe any concrete examples. I would and many companies do trust an LLM with low to medium skilled text processing tasks: internal knowledge sharing, first line customer support, low stakes document review. Github has their famous Copilot product, which I don't personally find too useful but many of my coworkers use.

In other words, you don't trust it with anything important.

Really depends on what you understand the word to mean. All of the things I listed are “important” in the sense that they’ve gotta get done, and in the sense that it matters how well they’re done. Would I trust an LLM with something business-critical? No, but there’s quite a lot of people I’d put in the same boat.

The difference is some humans exist that can be trusted with business critical tasks but no AIs exist with that level of competence.

This is actually a great example of what I'm talking about. All this talk about how useful AIs are, and how humans have flaws and yet the conclusion is always the same. AIs are only useful for tasks that are relatively easy and have a higher tolerance for failure.