Share with your CTO
Google’s Agent2Agent Protocol (A2A) has joined the Agentic AI Foundation (AAIF) as its fifth hosted project, sitting alongside Model Context Protocol (MCP) and three other open specifications under Linux Foundation governance. A2A solves a different problem than MCP: where MCP connects an agent to tools and data, A2A lets independent agents from different organizations discover each other and transact securely across vendor and framework boundaries. The AAIF now hosts the full protocol stack that enterprise multi-agent systems need to run at scale, with 150-plus organizations already backing A2A.
What this means for your business
The question of which multi-agent protocol wins matters less right now than the question of whether your architecture assumes agents will stay inside your own security perimeter. Most current enterprise agent deployments do make that assumption, and it’s already wrong. The moment an agent needs to transact with a supplier, a bank, or a logistics partner, you’ve crossed into A2A territory whether you planned for it or not. Where you sit on this depends on how many of your agent use cases touch external counterparties.
The APAC adoption pattern in the article deserves more attention from Western enterprise architects than it’s currently getting. Huawei has already standardized A2A as the protocol between its OS-level assistant Celia and in-app agents across HarmonyOS, and Tencent’s WeChat is integrating over it with dual authorization. That isn’t a proof of concept. It’s a production deployment at consumer internet scale, and it validates the cross-boundary handoff model that Western financial services firms are now starting to pursue. Visa and Wells Fargo joining the AAIF on August 13th isn’t incidental, it’s a directional signal that B2B2C agent transactions are moving from whiteboard to procurement.
The governance maturity gap between MCP and A2A is real and worth tracking. MCP just completed its largest revision since launch, moving to a stateless architecture compatible with standard load balancers and Kubernetes clusters and hardening its OAuth implementation, with a formal 12-month deprecation guarantee that enterprises can actually plan around. A2A has 150 backing organizations and a governance home, but hasn’t yet gone through that kind of production-hardening cycle. If your roadmap puts cross-organizational agent transactions into production inside 18 months, the protocol you’re betting on is still maturing. That’s not a reason to wait, but it is the specific risk to name in your architecture review.
Concept deep-dive: Agent-to-agent interoperability
A2A is the protocol layer that lets two AI agents, owned by different organizations with different security models, find each other and exchange work reliably. Think of it as the TCP/IP handshake for autonomous software agents: before two services on the internet can communicate, they need a shared language for establishing identity and passing data. A2A provides that for agents. The business consequence is that any workflow crossing an organizational boundary, a payment, a booking, a supply chain event, needs this layer to be auditable and secure.
Based on reporting from What Google’s A2A joining the Agentic AI Foundation means for enterprise agent architecture, originally published 2026-09-10 15:46:00.
