By the end of 2026, more than 250 companies, including Google, Microsoft, Amazon, Anthropic, OpenAI, Bloomberg, Shopify, and Block, will belong to a single foundation built around one question: can AI agents from different vendors actually talk to each other? AI agent interoperability just took its biggest step yet. On August 17, 2026, Google confirmed that its Agent2Agent Protocol (A2A), the open standard that lets independent AI agents discover one another and coordinate tasks, is moving into the Agentic AI Foundation (AAIF), joining Anthropic’s Model Context Protocol (MCP) under the same neutral roof. For any business running agents from more than one vendor, and most now are, this shift matters more than it sounds. This article breaks down what changed, why the Agentic AI Foundation is becoming the center of gravity for agent standards, and what it means for how you build, buy, and connect AI agents going forward.

What A2A’s Move Means for AI Agent Interoperability

Until now, A2A lived inside the Linux Foundation’s broader project portfolio, one open source initiative among hundreds. Moving it into the Agentic AI Foundation puts it in the same house as MCP, the protocol that connects individual AI agents to their tools and data sources. The distinction matters: MCP handles what an agent can access, while A2A, the AI agent communication protocol at the center of this move, handles how independent agents find each other and hand off work. Google Cloud VP Rao Surapaneni told Axios that the original idea behind A2A was simple: enterprises were already deploying agentic systems from multiple technology providers, and those systems needed a common way to work together rather than requiring custom point-to-point integrations for every pairing. Housing A2A and MCP under one foundation is meant to close that gap. Instead of a patchwork of one-off connectors between every agent platform a company uses, the goal is a shared, model-agnostic layer that lets teams pick providers based on cost, performance, or latency rather than which integrations happen to exist. AAIF executive director Mazin Gilbert framed it as a distinction between an open protocol and an open standard: a single open protocol is useful, but real interoperability requires the whole stack, identity, discovery, messaging, and tool access, to be open and working together. That is the bet the Agentic AI Foundation is now making.

Inside the Agentic AI Foundation’s Rapid Growth

The Agentic AI Foundation did not exist until December 2025, when the Linux Foundation launched it with founding contributions of MCP, Block’s goose, and OpenAI’s AGENTS.md. Less than a year later, it has grown from fewer than 40 members to more than 250, a pace AAIF says outstrips the Cloud Native Computing Foundation’s growth at the same stage. That roster now includes nearly every major AI lab and cloud provider, Google, Microsoft, Amazon, Anthropic, OpenAI, Bloomberg, Cloudflare, Shopify, and Block among them, an unusually broad coalition for companies that compete directly on agent platforms. The foundation’s flagship events, AGNTCon and MCPCon, are scheduled for North America and Europe later in 2026, signaling an attempt to build the same kind of ecosystem gravity that Kubernetes achieved for cloud infrastructure a decade earlier. For a market this young, that speed of adoption is notable. It suggests vendors have concluded that fragmented, incompatible agent protocols would slow enterprise adoption for everyone, not just smaller players. Rather than each major lab pushing its own closed standard, the biggest names in agentic AI are now, at least on paper, co-governing the plumbing that lets their agents interoperate. Whether that cooperation holds once real commercial incentives collide is the open question, but the membership numbers alone mark 2026 as the year agent standards stopped being a side project and became core infrastructure.

How AI Agents Communicate With Each Other: Interoperability in Practice

So how do AI agents communicate with each other under this emerging stack? In practical terms, A2A gives agents a shared way to publish what they can do, discover other agents with relevant capabilities, and negotiate how to exchange information, whether that’s plain text, structured forms, or media, without exposing their internal memory, models, or proprietary logic. MCP, working alongside it, governs how each individual agent reaches out to tools, files, and data sources it needs to complete a task. Together they cover both halves of an agentic workflow: what an agent can do internally, and how it works with other agents externally. For a business, the practical takeaway is straightforward. If you are evaluating AI agent platforms in 2026, ask vendors directly whether they support A2A and MCP rather than a proprietary equivalent. Standards-based agents are far cheaper to connect to your existing stack, and they protect you from being locked into a single provider as pricing and capabilities shift. Teams already juggling AI agent sprawl, running a dozen or more disconnected agents across departments, should treat protocol support as a procurement requirement, not a nice-to-have. It is also worth auditing any custom integrations your team has already built between agent systems; some of that glue code may become unnecessary as standards-based connections mature over the next few quarters, especially following the recent MCP specification update.

What’s Next for Open AI Agent Standards

This move does not settle the interoperability question, it accelerates it. A2A and MCP now share a governance home, but dozens of narrower protocols, from payment rails like x402 to identity frameworks for agent credentials, still need to find their place in the same open stack. Gilbert’s comment that companies want the whole stack open, not just one protocol, points to where AAIF is likely headed next: folding in adjacent standards for agent identity, memory, and cross-network payments rather than stopping at communication and tool access. There is a reasonable case for skepticism here too. Standards bodies with 250 competing members can move slowly, and past attempts at unifying fast-moving technical ecosystems have sometimes produced compromise specifications that satisfy no one fully. Still, the fact that Google, Microsoft, Amazon, Anthropic, and OpenAI are all backing the same neutral foundation, rather than each pushing a competing standard, is a meaningfully different starting position than most infrastructure wars begin from. If the coalition holds, 2026 may be remembered as the year AI agents stopped being isolated islands and started forming a genuinely interoperable network.

Key Takeaways

Three things are worth remembering from this shift. First, A2A joining the Agentic AI Foundation puts agent-to-agent communication and tool access (MCP) under one roof, reducing the custom integration work businesses face when mixing vendors. Second, the foundation’s growth from 40 to over 250 members in under a year signals unusually broad industry buy-in for a standards effort this young. Third, if you are deploying or evaluating AI agents, protocol support (A2A and MCP compatibility) should now be a standard question in your procurement checklist, not an afterthought. Explore more breakdowns of the tools, platforms, and trends shaping agentic AI at BigAIAgent, from AI agent frameworks to the platforms managing agent sprawl. As open standards mature, will your organization pick agent platforms based on features alone, or will protocol compatibility become the deciding factor?

Sources: Axios, Linux Foundation