# Exploring Advanced LLM Multi-Agent Systems Based on Blackboard Architecture

**Authors:** Bochen Han, Songmao Zhang  
**arXiv:** [2507.01701](https://arxiv.org/abs/2507.01701)  
**Date:** 2025-07-02  
**Summary page:** https://www.summarizepaper.com/en/paper/2507.01701v1/

## Key points

- Authors Bochen Han and Songmao Zhang propose integrating blackboard architecture to enhance multi-agent systems (MASs)
- Integration enables collaborative sharing of information and messages among agents with diverse roles
- Selection of actions by agents is based on the current content of the blackboard, leading to consensus among all agents
- Experiments conducted using commonsense knowledge, reasoning, and mathematical datasets show system competes with state-of-the-art MASs with superior performance and fewer tokens
- Incorporating blackboard architecture into LLM MASs improves collaboration, decision-making, and overall performance
- Potential for facilitating complex problem-solving scenarios where traditional structures are not sufficient

## Summary

In their paper titled "Exploring Advanced LLM Multi-Agent Systems Based on Blackboard Architecture," authors Bochen Han and Songmao Zhang propose a novel approach to enhance multi-agent systems (MASs) by integrating the blackboard architecture. This integration allows agents with diverse roles to collaboratively share information and messages throughout the problem-solving process. The selection of agents to take actions is based on the current content of the blackboard, and this process is repeated until a consensus is reached among all agents. The authors developed the first implementation of this proposal and conducted experiments using commonsense knowledge, reasoning, and mathematical datasets. The results of these experiments demonstrate that their system can compete with state-of-the-art static and dynamic MASs, achieving superior average performance while also utilizing fewer tokens. This innovative approach has the potential to facilitate complex and dynamic problem-solving scenarios where traditional structures or workflows are not readily available. Overall, this research highlights the effectiveness of incorporating the blackboard architecture into LLM MASs, showcasing its ability to improve collaboration, decision-making, and overall performance in multi-agent systems. The findings suggest promising implications for advancing problem-solving capabilities in various domains where intricate tasks require coordinated efforts from multiple agents.