Modern recommender systems have undergone many innovations that have revolutionized the way they work and interact with users. Multi-agent systems, where multiple specialized agents work together to achieve complex goals, represent a promising approach for advancing recommender systems. Large Language Models (LLMs) can be considered agents of a multi-agent system composed of different types of agents. This research aims to explore the potential of multi-agent systems for recommendation, including both LLM agents as well as other possible kinds of agents, such as retrieval ones. Key exploration areas include the implications for the multistakeholder problem, addressing beyond-accuracy concerns, such as fairness and bias of recommendations, and tackling other issues, like the agent’s coherence during a dialog or the definition of an evaluation framework for the conversational agents. Currently, multi-agent systems are mainly based on cooperation between agents, but we will also explore the competition between them, which may help in finding a trade-off between accuracy and fairness or between different stakeholders’ objectives.