Coordination and Communication Standards for LLM-Based Multi-Agent Systems

Investigate coordination mechanisms for large language model-based multi-agent systems to determine how agents can reach consensus and allocate tasks among themselves, and identify standardized communication languages or message formats that enable efficient and unambiguous information exchange among heterogeneous agents.

Background

The paper discusses emerging multi-agent AI ecosystems where language-model-based agents collaborate on complex tasks. While natural-language interaction among agents is flexible, it can be inefficient or ambiguous without structured protocols.

The authors explicitly note open questions regarding both the mechanisms by which agents coordinate (consensus and task allocation) and the communication formats they should use to exchange information. Historical agent communication languages (KQML, FIPA-ACL) provided structure but were not widely adopted in modern LLM-based contexts, highlighting the need for new, practical standards.

References

There are open questions in research about coordination strategies (how do agents reach consensus or allocate tasks among themselves?) and communication protocols (what language or format should agents use to exchange information efficiently and without ambiguity?).

— The Coral Protocol: Open Infrastructure Connecting The Internet of Agents  (2505.00749 - Georgio et al., 30 Apr 2025) in Section 2, Subsection "Multi AI Agent Collaboration"

Discovered protocols also assume centralized access to all agent states; whether effective protocols can be discovered under partial observability is an open question.

— Discovering Adaptive Transmission Programs for Collective Innovation  (2608.24545 - Colas et al., 25 Aug 2026) in Section 4, Discussion, paragraph “Limitations”