AWS Machine Learning shows an A2A reference architecture for cross-organization agents

In an AWS Machine Learning blog post, AWS describes a deployed reference architecture for cross-organization agent collaboration using the A2A protocol. A LangChain-based enterprise orchestrator communicates with a Strands Agents SaaS agent across separate AWS accounts, with OAuth 2.0 and OIDC handling authentication on both sides. The post says the system can be provisioned and tested from three Jupyter notebooks, and that new skills are added by registering Agent Cards without changing agents already running.