Multi-Agent Systems

Multi-agent systems (MAS) in the context of AI refer to systems where multiple autonomous AI agents collaborate to accomplish complex tasks that exceed the capabilities of individual agents.

Architecture patterns

Common MAS architectures include:

• Hierarchical – Coordinator agent delegates to specialized workers
• Peer-to-peer – Agents collaborate as equals
• Marketplace – Agents advertise capabilities and negotiate tasks

Communication protocols

Standardized communication is essential for MAS:

• Agent-to-Agent Protocol – Google's enterprise standard
• Agent Communication Protocol – IBM's local-first approach
• Agent Network Protocol – Decentralized internet-scale communication

See also

• Agent-to-Agent Protocol
• Agentic Web