Isn't this kind of the hope/dream of multi-agent systems where one LLM "coordinates" among others or checks the responses? In my experience it works about as w

Isn't this kind of the hope/dream of multi-agent systems where one LLM "coordinates" among others or checks the responses? In my experience it works about as well as you're describing.