Multi-Agent Collaboration Mechanisms: A Survey of LLMs

Source Context

A comprehensive survey from University College Cork and collaborating institutions that provides a framework for understanding how LLM-based agents collaborate in multi-agent systems.

Core Framing

The authors argue that while LLM-based multi-agent systems (MAS) are proliferating, existing surveys focus on applications and architectures without adequately addressing the mechanisms of collaboration itself. Their contribution is a five-dimensional framework characterizing collaboration by: actors, types, structures, strategies, and coordination protocols.

Key Contribution

A taxonomic framework that decomposes MAS collaboration into analyzable dimensions, enabling systematic comparison of existing approaches and identification of research gaps.

Extracted Content

Atoms

• 05-atom—collaboration-channel-definition
• 05-atom—horizontal-vs-vertical-scaling
• 05-atom—coopetition-definition
• 05-atom—collaboration-type-taxonomy
• 05-atom—collaboration-strategy-taxonomy
• 05-atom—communication-structure-taxonomy
• 05-atom—static-vs-dynamic-orchestration
• 05-atom—competition-improves-outputs
• 05-atom—theory-of-mind-in-mas

Molecules

• 05-molecule—mas-collaboration-framework
• 05-molecule—role-based-agent-specialization

Why This Matters

As agentic AI systems become more prevalent, the question shifts from “can multiple agents work together?” to “how should they work together?” This framework provides vocabulary and conceptual structure for making those design decisions.

Related:, 05-atom—agentic-ai