Agent Communication (Multi-Agent Systems)

A single agent can only do so much. Multi-agent systems split work across several specialised agents — but for them to work together, they must communicate. Agent communication is the substrate of every multi-agent system: how agents exchange tasks, results, and context. Get it right and agents cooperate; get it wrong and they talk past each other.

💡 In one line: Agent communication is how agents exchange messages — tasks, results, and context — so they can coordinate.

What is a Multi-Agent System?

A multi-agent system is several agents — each with its own role and tools — working together on a task. Instead of one generalist, you have specialists (a researcher, a writer, a reviewer) that coordinate by communicating.

What is Agent Communication?

It's the set of mechanisms by which agents exchange information: passing tasks, results, questions, context, and status between one another. It's the channel that makes coordination possible.

Why It Matters

Communication is the foundation for the rest of this section — collaboration, delegation, and orchestration all ride on it. Without a reliable way to exchange messages, agents simply can't coordinate.

Communication Patterns

  • Direct messaging — one agent sends straight to another.
  • Via an orchestrator / router — a coordinator relays messages.
  • Broadcast — one agent messages many.
  • Shared memory / blackboard — agents read and write a common state.

Message Anatomy

A message carries a sender, receiver, content, and intent.

Whiteboard
Whiteboard diagram


 Protocols & Formats

Messages can be structured (a role + content object), natural language, or structured data. Standards are emerging — Agent2Agent (A2A)-style protocols for agent-to-agent communication, and MCP for connecting agents to tools and data.

Synchronous vs. Asynchronous

  • Synchronous — the sender waits for a reply (simple, but blocking).
  • Asynchronous — the sender continues and handles replies later (scalable, but more complex).

Challenges

  • Message overload and token cost — every exchange consumes context.
  • Miscommunication and coordination errors or loops.
  • Keeping shared state consistent across agents.

Best Practices

  • Use structured messages with clear roles and intent.
  • Limit chatter to control cost.
  • Use shared memory for common context.
  • Define a protocol so agents agree on the format.

Summary

  • Agent communication is how agents exchange tasks, results, and context.
  • Patterns include direct, orchestrated, broadcast, and shared-memory.
  • Messages carry sender, receiver, content, and intent.
  • Standards like A2A and MCP are formalising it.
  • It's the foundation for collaboration, delegation, and orchestration — next. EOF echo created