liveBook · Manning

Agent communication

As explained in AI Agents in Action, Second Edition

Agent communication is the exchange or sharing of information among agents in a software system.

Agent communication is the exchange of information among agents in a software system. It determines which context each agent can access and how agents interact while completing a task. section 4.1 section 4.1.2

Communication is distinct from decision-making and control. Decision-making chooses the next action, control executes it, and communication carries information between agents. section 4.1

Choose a communication pattern according to the context each agent needs:

  • Message passing sends only the required messages or objects, reducing noise and context cost; its quality depends on the previous agent's output. section 4.1.2
  • A shared message thread gives every agent the previous exchanges and can support recovery, but growing context increases token use and may reduce focus. section 4.1.2
  • Tool exchange through MCP provides structured, narrowly scoped communication, but adds the overhead of a tool call for each exchange. section 4.1.2

Competitive analysis

An orchestrator assigns one research agent to each of three companies. Each researcher receives only its company name, rubric, and output schema. A synthesis agent receives the three completed reports, but not the orchestrator's internal planning notes. This boundary avoids unnecessary cost and helps each agent focus on relevant evidence. section 4.1.2

Table 4.3 Flow communication patterns: Pros and cons

Communication pattern

Pros

Cons

Message passing

Simple and effective

Limits messages to just what is needed

Filters out noise

Messages are only as good as the previous agent’s output.

Shared message thread

All agents are aware of previous exchanges.

Robust; failures in agent execution may be recoverable

Context window size grows with each agent. Increases token use.

Can cause confusion; agents may lose focus and perform other tasks

Tool exchange (MCP)

Limits messages to just what’s needed

Communication is structured and well-defined.

Overhead of making a tool call for each communication

Compare message passing, shared message threads, and MCP by their benefits and failure modes. section 4.1.2

How AI Agents in Action, Second Edition covers it

How deep it goes
Explains three communication strategies with costs, selection risks, and a scoped competitive-analysis example.
Examples
Uses parallel company research, synthesis, message threads, message passing, and MCP as concrete communication examples.
What it assumes
Assumes basic awareness of agents while defining decision-making, control, and communication from first principles.
What you can do after
Evaluate communication patterns and scope agent context to balance cost, relevance, robustness, and implementation overhead.
How it connects
Connects multi-agent systems with decision-making, control, shared memory, message passing, MCP, and orchestration.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of AI Agents in Action, Second Edition. The section links point to the parts of the book it draws on.