AAMAS 2026 Tutorial

A Concise Introduction to LLM-based Multi-agent Systems

Presenters: Yang Chen, Shuyue Hu
Conference: AAMAS 2026, Paphos, Cyprus
Date: May 25-29, 2026 · Format: Half-day tutorial (3.5 hours technical content)

Official AAMAS 2026 Website

Overview

This tutorial provides a concise, agent-centric introduction to LLM-based multi-agent systems (MAS), with an emphasis on the interplay between large language models and classical multi-agent principles. We discuss how multi-agent mechanisms such as debate, role specialization, coordination protocols, and strategic interaction improve reliability and robustness beyond single-agent LLM settings, and how LLMs can serve as infrastructure for communication and coordination in MAS.

Detailed Outline

Part I: Foundations of LLMs and LLM-based Agents

1 hour

Agent abstraction: policy, memory, and interaction loop
Single-agent LLMs as decision-making entities
Roles of LLMs in agents: priors, planners, and reasoning modules
From single-agent to multi-agent systems

Break: 10 minutes

Part II: Multi-Agent Systems for LLMs

1 hour

Limitations of single LLM agents: hallucination, myopia, bias, and brittle reasoning
Debate, deliberation, and critique for improved reasoning reliability
Redundancy, voting, and consensus for robustness

Break: 10 minutes

Part III: LLMs for Multi-Agent Systems

1 hour

Advantages of LLMs as infrastructure for MAS
Language-mediated coordination and communication
Connections to mechanism design, learning in games, and collective decision-making

Break: 10 minutes

Part IV: Open Challenges and Future Directions

30 minutes

Evaluation of interactive, deliberative, and strategic agent systems
Scalability in agent number, interaction length, and reasoning depth
Efficient communication in LLM-based agents
Open theoretical questions at the intersection of MAS and LLMs

Audience and Prerequisites

The tutorial is designed for researchers, practitioners, and advanced graduate students in multi-agent systems, machine learning, and AI.

Basic familiarity with agents and multi-agent systems Introductory machine learning or reinforcement learning No prior LLM experience required

Presenters

Yang Chen

Shanghai AI Lab, Shanghai, China

Research focus: reinforcement learning for LLMs, reward modeling, and LLM-based multi-agent systems. Active contributor to the AAMAS community.

Email: [email protected]
Website: yangchen.info

Shuyue Hu

Shanghai AI Lab, Shanghai, China

Research focus: multi-agent systems, LLMs, and game theory, including multi-agent debate, routing, and emergent behaviors in LLM-based MAS.

Email: [email protected]
Website: shuyuehu.github.io

Contact

Corresponding presenter: Yang Chen

[email protected] · [email protected]