Agent Communication

Introduction

When agents talk to each other (or to humans), use a structured protocol: JSON messages with type, sender, payload. Free-form chat between agents creates ambiguity and runaway loops.

Beginner analogy: radio operators use call signs and codes, not casual chat — same reason.

Understanding the topic

Core concepts:

• Define message types: REQUEST, RESPONSE, ERROR, STATUS.
• Use a queue (Redis, SQS, Kafka) for async comms.
• Always include a correlation ID.
• Time out conversations — agents shouldn't chat forever.
• Log every message for replay/debug.

Syntax reference

Visual workflow / architecture:

┌────────┐  REQUEST  ┌────────┐
│ Agent A│──────────►│ Agent B│
└────────┘           └────┬───┘
                          │ RESPONSE
                          ▼
                     ┌────────┐
                     │ Agent A│
                     └────────┘

Real-world use

AutoGen, CrewAI and OpenAI Swarm define explicit message shapes; production deployments add queues.

Best practices

• Treat agent comms like microservice APIs.
• Enforce schemas with JSON Schema / Zod.

Common mistakes

• Letting agents chat in free text — they spiral into pleasantries.

Hands-on exercise

Interview preparation — practice these questions:

• Q1. Why use structured protocols between agents?
• Q2. Three message types you'd define.
• Q3. Scenario: two agents loop forever exchanging 'are you sure?'. Fix?