Reasoning Capacity in Multi-Agent Systems: Limitations, Challenges and Human-Centered Solutions