I do some experiments with what I can run on my 4090 GPU. Some things are fine. Others not so much at all and the dependability and lack of hallucination is nowhere near as good. So I don't agree it is straightforward to get there with local AI. Currently I am dependent on open source models run through nano-gpt when I need heavier lift.