AWS ships reference architecture for cross-org multi-agent systems over A2A
AWS ML Blog · rss · 2026-11-12
AWS ML Blog published a deployable reference architecture for composable multi-agent systems across organizational boundaries using the Agent-to-Agent (A2A) protocol, now a Linux Foundation project.
Key design
- A LangChain enterprise orchestrator and a Strands-based compliance SaaS agent live in separate AWS accounts, communicating solely via A2A — no shared secrets, no framework dependency on either side.
- The orchestrator fetches the SaaS agent's Agent Card at startup and auto-generates a LangChain tool per skill, so the model decides routing with no hardcoded logic.
- The compliance agent calls back the enterprise context agent for metadata (production accounts, policies, escalation contacts), making agents true peers.
Security & protocol
- Bidirectional OAuth with separate Cognito user pools per side; onboarding exchanges only public values (IdP URLs, client IDs, scopes).
A2A vs MCP: MCP connects an agent to its own tools (client-server); A2A connects agents as peers without exposing internals. In production, the SaaS agent could use MCP internally and expose itself via A2A.
Orchestrator and compliance agent run Claude Sonnet 4.6; context agent uses Amazon Nova Lite. Deployed from three Jupyter notebooks.
More from coding & agent
- Demo: Atlas + Fable 5.2 + three.js for 3D Scene Generation — XRarchitect · 2026-11-13
- DocFS turns documents into file systems so VLM agents can reason over them — _reachsumit · 2026-11-13
- Indie devs are ditching Intercom: "you don't use 90% of the product" — jonathan_wilke · 2026-11-13
- Pinokio dev ships Zonos3-Mini Voice Studio with realtime voice cloning on Mac — cocktailpeanut · 2026-11-13
- Pinokio apps now reachable from any agent: ask Claude to clone a voice — cocktailpeanut · 2026-11-13
- Gemini 4 Hydrogen lands in Cursor, scoring 54.3% on CursorBench vs 39.6% for 3.8 Flash — mattyp · 2026-11-13