LLMs aren't for complex reasoning. LLMs are capable of doing associative tasks in a given context. System 2 can be implemented in Agents concept (think slow - chain of simple steps).