Two Protocols, One Governance Home
The two most important connective standards in agentic AI now live under one roof. On August 17, 2026, Google’s A2A protocol joined the Agentic AI Foundation as a hosted project, joining Anthropic’s Model Context Protocol, which was one of AAIF’s founding contributions. The Agentic AI Foundation itself is a young body: the Linux Foundation announced its formation on December 9, 2025, anchored by three founding project contributions — MCP from Anthropic, goose from Block, and AGENTS.md from OpenAI.
That founding roster already signalled something unusual: three companies that compete directly for enterprise AI budgets each donated a piece of core infrastructure to a neutral steward rather than keeping it proprietary. The A2A addition extends the pattern to the fourth major standard in the stack, and it does so with heavyweight backing. The foundation’s platinum members include Amazon Web Services, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft and OpenAI — the same set of firms whose commercial rivalry defines the current AI market.
The scale of the buy-in is what makes the move notable. AAIF grew from fewer than 40 members at its December 2025 launch to more than 250 within about eight months, according to the foundation’s statement reported by Axios. For a standards body barely two quarters old, that trajectory reflects how quickly the industry concluded that agent interoperability is a shared-infrastructure problem, not a competitive battleground — a conclusion the web itself reached decades ago with HTTP and TCP/IP, and the container ecosystem reached with Kubernetes.
What A2A and MCP Actually Do — and Why They Are Different
The two protocols are complementary, not competing, and understanding the split is the key to why housing them together makes sense. MCP standardizes how a single agent reaches out to the resources it needs — a database, an API, a file system — so that a tool built once can be consumed by any MCP-aware agent. A2A works on a different axis: it standardizes how one agent asks another agent to complete a task and return the result, even when those agents were built by different vendors and run inside different organizations.
Put simply, MCP operates vertically — agent to resources — while A2A operates horizontally — agent to agent, across organizational boundaries. MCP first appeared in November 2024 as Anthropic’s contribution to the ecosystem; A2A launched in April 2025 under Google. Google had already donated A2A to the Linux Foundation earlier, so the August move clarifies where it is hosted and governed rather than transferring control for the first time. The practical upshot is that a company can now build agents that both pull in external data (via MCP) and delegate work to third-party agents (via A2A) using two openly-governed standards rather than a stack of vendor-specific integrations that break every time a supplier changes its API.
The Fragmentation Risk the Move Is Meant to Prevent
The reason competitors cooperate here is defensive as much as idealistic. In the absence of shared standards, every major AI vendor tends to build its own agent-communication layer, and enterprises end up locked into whichever ecosystem they adopted first. That fragmentation is expensive: integration teams spend their time writing and maintaining brittle adapters between incompatible agent frameworks instead of building the automation that actually creates value. A neutral foundation reduces that tax by giving every vendor a credible commitment that the core protocols will not be quietly changed to favour one player.
The governance detail matters more than it might appear. A protocol controlled by a single company — even a well-intentioned one — carries the permanent risk that the company will eventually optimize the standard for its own commercial interests, or deprecate it when priorities shift. Housing A2A and MCP inside a Linux Foundation body puts change control in the hands of a neutral steward with a long track record of running exactly this kind of multi-vendor infrastructure, from Linux itself to Kubernetes. For enterprises making multi-year bets on agentic architectures, that neutrality is the difference between building on bedrock and building on a vendor’s roadmap.
Advertisement
What This Means for Enterprises Building on AI Agents
1. Treat A2A and MCP support as a procurement filter, not a nice-to-have
When evaluating agent platforms and AI vendors, ask directly whether they support MCP for tool access and A2A for inter-agent communication. A vendor that only speaks its own proprietary protocol is a lock-in risk; one that implements the AAIF-governed standards gives you an exit path and interoperability with the rest of your stack. Make protocol support an explicit line item in RFPs from this procurement cycle onward.
2. Design agent architectures around the vertical/horizontal split now
Separate the two concerns in your own designs: use MCP-style connections for how agents reach data and tools, and reserve A2A-style delegation for how agents hand tasks to one another. Even if you are early in your agent journey, architecting to this split today means you can swap individual components later without re-plumbing the whole system when the standards mature.
3. Track the governance, not just the spec — and budget for change
Neutral governance does not mean the protocols are frozen; it means changes go through a transparent multi-vendor process. Assign someone on your platform team to follow AAIF’s project roadmaps so you are not surprised by a breaking change, and budget for periodic protocol upgrades the same way you budget for framework version bumps. The upside of open governance is visibility into what is coming; capture it.
The Bigger Picture: Standards Are the Real Moat Now
The consolidation of A2A and MCP under one neutral foundation is a signal about where the agentic AI market is maturing. When rivals as fierce as Google, Microsoft, OpenAI, Anthropic and AWS agree to steward shared plumbing together, it usually means the competitive frontier has moved up the stack — away from who controls the connective protocols and toward who builds the most useful agents on top of them. That is a healthy sign for buyers: it suggests the plumbing is becoming a commodity, which lowers switching costs and raises the odds that the agents an enterprise builds this year will still interoperate with the ecosystem in three years. The open question is execution — whether AAIF can keep 250-plus members with directly conflicting commercial interests aligned as the standards evolve. History with the Linux Foundation’s other projects suggests it can, but the stakes here are higher and the market is moving faster than any it has governed before.
Frequently Asked Questions
What is the Agentic AI Foundation and when was it created?
The Agentic AI Foundation (AAIF) is a Linux Foundation body whose formation was announced on December 9, 2025. It hosts and governs core agentic-AI standards under neutral, multi-vendor control. Its founding project contributions were Anthropic’s Model Context Protocol (MCP), Block’s goose, and OpenAI’s AGENTS.md. It grew from fewer than 40 members at launch to more than 250 within about eight months.
How are A2A and MCP different?
They solve different problems. MCP standardizes how a single agent reaches resources like databases, APIs and file systems (a vertical, agent-to-resources connection). A2A standardizes how one agent asks another agent to complete a task and return the result (a horizontal, agent-to-agent connection across organizational boundaries). They are complementary, not competing.
Why does neutral governance of these protocols matter for enterprises?
A protocol controlled by a single company can be changed to favour that company’s commercial interests or deprecated when its priorities shift. Housing A2A and MCP inside a Linux Foundation body puts change control with a neutral steward, giving enterprises a credible commitment that the standards their agent architectures depend on will evolve through a transparent multi-vendor process rather than one vendor’s roadmap.
Sources & Further Reading
🔗 Related Intelligence
From Agents to AI Operating Systems: The Next Layer of Computing
MCP: How the Model Context Protocol Is Becoming the USB-C of AI Integration
AI Agents: The Complete Intelligence Hub
AI Tool Ecosystems: How Agents Connect to Software
