> The idea is simple in theory

Unfortunately it is very difficult in practice. And a wise person once said: “Theory and practice sometimes clash. And when that happens, theory loses. Every single time.”

The problem will become thorny when you consider what can be automated vs. what can't be, and what should be vs. what shouldn't. IOW, what do you consider a "routine decision".. after all it is subjective.

It is completely unreasonable to expect that people will give you their money and then never want to control it in some way. In fact, it's a contradiction really.

So I certainly applaud you for considering the problem and more so if you make an attempt at some implementation. However, it is a bit like asking "will AI become self-aware and make humans obsolete". For the foreseeable future people will not relinquish the ability to somehow "pull the plug".

Short answer: Never

Why you might think this good idea. LLMS are great at discussing best practices. I could probably improve my financial health by 20% by asking for and following best practices advice.

Why I’d never put it in control: LLMs are shit at doing things right the first time, dealing with the unexpected, dealing with the real world.

If you haven’t had an LLM make catastrophic, terrible and insane mistakes you haven’t been using it enough to see how it can go wrong and have no business either making or using AI for anything important.