Multi-Agent Large Language Models for Conversational Task-Solving