A2A in 45 Minutes: Building Interoperable AI Agents
summary
A2A is an open, Google‑origin spec now at the Linux Foundation that gives agents discovery, secure messaging and coordinated tasks with status (submitted, working, input_required, completed, failed, canceled, rejected, authorization_required, unspecified) over HTTP/SSE/JSON‑RPC. Demo shows Java (Quarkus + LangChain4J A2A 1.0 beta) and a Python agent exchanging multimodal artifacts and negotiating work; an AP2 payment extension exists. It runs end‑to‑end but the ecosystem is young and migrations from 0.3→1.0 and SDK gaps remain.
AI-written summary · about
chapters (16)
- 0:00Introduction and speaker intro
- 1:55What is an AI agent
- 3:05Why A2A is needed
- 5:37A2A basics and features
- 8:24Agent cards explained
- 10:33Task workflow and statuses
- 18:43A2A vs MCP comparison
- 20:46Code demo setup
- 23:16Building agent card in Java
- 24:21Agent executor implementation
- 26:54Handling task failures
- 27:56Client-side A2A interaction
- 30:23Live demo execution
- 32:13Cross-language agent example
- 33:41A2A ecosystem and future
- 36:52Resources and Q&A
related talks
43mA2A in 50 Minutes: Building Interoperable AI Agents
27mA2A in 25 Minutes: Building AI agents that communicate
31mA2A in 30 Minutes: Building Communicating AI Agents
34mWhat can A2A do for you and what can you do for it
16mAgent2Agent Protocol and the Future of Agent Collaboration
10mKeynote: The Agent2Agent (A2A) Protocol
02h 59mStandards and AI Agents: an Overview of MCP, A2A, ADK and More
46mMCP and A2A Revolutionizing Interconnection and Collaboration of AI Agents