Module Overview
This module covers what it takes to observe agents in production and have them talk to each other across frameworks. We pair the operational layer (tracing and metrics) with the interoperability layer (A2A).
Learning Objectives
- Instrument agents with production observability tooling.
- Explain the A2A protocol: Agent Cards, the task lifecycle, and transport.
- Build an A2A-compliant agent server and a client that can discover and delegate to it.
- Decide between MCP and A2A — and use them together when needed.
Topics Covered
Agent Observability Tools
- LangSmith
- Logfire
A2A (Agent-to-Agent) Protocol
- Why A2A exists — the multi-agent interoperability problem
- A2A architecture — agents, Agent Cards, and the task lifecycle
- Agent Cards — capability discovery and the well-known endpoint
- A2A task model — submitted, working, completed, failed, and streaming states
- A2A transport — HTTP, SSE, and JSON-RPC message structure
- Building an A2A-compliant agent server
- Building an A2A client — discovering and delegating to remote agents
- A2A vs MCP — when to use which and how they complement each other
Key Concepts & Terminology
Agent Card, well-known endpoint, task lifecycle states, JSON-RPC envelope, server-sent events.
Tools & Frameworks Referenced
LangSmith, Logfire, A2A protocol implementations.
Prerequisites
Modules 20–21 (function calling, MCP, LangGraph).