Skip to content
ATLACIS

AI Decision Support

Google and Anthropic's AI agent standards now share one governance home. Here is what business owners should know before trusting a vendor's 'open' claim.

On August 17, 2026, Google's Agent2Agent (A2A) protocol, the open standard that lets one AI agent hand off work to another, became a hosted project of the Agentic AI Foundation (AAIF), the Linux Foundation body that already governs Anthropic's Model Context Protocol (MCP), the standard that lets an agent reach outside tools and data. The direct answer for a business owner: this is a governance change, not a new product or a new capability. Nothing about how these protocols work changed on this date. What changed is who is responsible for keeping them neutral as more vendors build agent products on top of them, and that matters because "built on an open standard" is quickly becoming a claim every AI vendor makes, whether or not it is meaningfully true for their specific product.

By Fabio Rabelo · Founder, ATLACIS ·

Book a CallSee How It Works

What happened

On August 17, 2026, the Agentic AI Foundation announced that A2A, Google's protocol for agent-to-agent communication, joined the foundation as a hosted project, placing it alongside MCP, Anthropic's protocol for connecting an agent to outside tools and data. Google donated A2A to the Linux Foundation back in April 2025 with backing from AWS, Cisco, Microsoft, Salesforce, SAP, and ServiceNow. This move is the protocol's second governance step, shifting it specifically into the newer Agentic AI Foundation the Linux Foundation set up in December 2025 to house agent-focused projects, MCP among them. Rao Surapaneni, Google Cloud's VP and general manager for business application platforms, said the move gives enterprises "the freedom to create more ambitious agentic systems" without being locked into one company's approach. AAIF's own leadership described it as filling in the last major piece of an already-published map of the open agent stack: MCP for agent-to-tool access, A2A for agent-to-agent communication, plus smaller pieces like AGENTS.md for coding-agent instructions. The detail worth knowing: nothing in the protocol itself moved. A2A's stable specification, version 1.0, shipped back in March 2026, and the underlying code repository, maintainers, and technical steering committee, with seats from AWS, Cisco, Google, IBM Research, Microsoft, Salesforce, SAP, and ServiceNow, are unchanged. What changed is which nonprofit foundation holds the trademark and the neutral paperwork, so the two most widely used agent protocols on the market now answer to the same governance body instead of two separate homes. AAIF itself has grown from 49 founding members in December 2025 to more than 250 today, with AWS, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft, and OpenAI among its platinum-level members.

Why it matters for business owners

Most business owners will never read a protocol specification. What they will see, increasingly, is a vendor's product page claiming their AI assistant, agent platform, or automation tool is "MCP compatible" or "built on open agent standards." Those labels are becoming a real purchasing signal, the same way "works with Zapier" or "API available" became one for an earlier generation of business software. This governance consolidation is the industry's attempt to keep that signal meaningful: instead of several competing, incompatible ways for one company's agent to talk to another company's agent, the two dominant approaches now sit under one neutral organization that no single company controls. That matters for cost and flexibility. If the standards stay genuinely open and interoperable in practice, a business can, in theory, mix agent tools from different vendors, Google's, Anthropic's, Microsoft's, or a smaller specialist's, without rebuilding a custom integration every time it wants to switch or add a tool. That is the opposite of the lock-in pattern that has made switching AI vendors slow and expensive for many businesses over the past two years.

What owners should not misunderstand

This was a governance move, not a technical one. The specification did not change, no new version shipped alongside the announcement, and existing implementations built against A2A or MCP need no update. A vendor cannot point to this news and claim their product suddenly gained interoperability it did not have the week before. It also does not mean every vendor's "supports A2A" or "MCP compatible" claim means the same thing. Two products can both implement the same protocol correctly and still fail to work together in practice if one adds proprietary extensions, implements only part of the specification, or leaves out the parts of the standard that actually matter for a specific workflow. Shared governance guarantees the rulebook is written by a neutral party. It does not audit or guarantee that any specific vendor's product actually follows that rulebook completely, or that two "compatible" products were ever tested together. And open governance does not mean free of vendor influence. The companies with seats on A2A's technical steering committee, and the platinum members funding AAIF, are largely the same cloud providers and model labs a business is choosing between in the first place. Open governance reduces the risk of one company unilaterally changing the rules later. It does not remove the commercial interests of the companies still writing the specification.

The operational lesson

A protocol becoming an open, neutrally governed standard shifts risk. It does not eliminate it. Before this consolidation, a business risked building on a protocol one company could change or abandon unilaterally. After it, that specific risk is lower, since no single company now controls either standard outright. But the risk moves down a level, from "which protocol do I build on" to "how completely does this specific vendor's product actually implement the protocol, and has anyone verified it works with the other tools I already use." This is close to the same lesson that showed up when Stripe agreed to acquire OpenRouter, the AI model gateway marketed as preventing vendor lock-in: a layer built to reduce dependency on one company can still end up owned, priced, or governed in ways a business did not choose. Open governance over a protocol is a stronger structural protection than one company's marketing promise, but it is still worth verifying rather than assuming.

What a serious business should do next

When a vendor's product claims to support MCP, A2A, or "open agent standards," ask which version of the specification they implement, whether they have tested actual interoperability with a second, independent tool rather than just their own suite, and whether they add proprietary extensions that would make switching away from them harder than the open standard implies. Do not treat "open standard" language on a product page as proof of portability. Ask for a specific example of it working with a competitor's tool. If your business already uses more than one AI agent tool, or is planning to, this is a reasonable moment to ask each vendor directly where they stand on MCP and A2A adoption specifically, since both protocols now sit under a governance structure every major lab has agreed to work through. That answer, more than a marketing claim, is what actually tells you how much flexibility you are buying.

The Atlacis view

An industry standard landing on neutral, shared ground is good news, but it is not a substitute for checking what a specific vendor actually built. Atlacis helps owners cut through "open standard" and "interoperable" language on a vendor's product page, verify what a specific AI tool actually supports before it gets wired into a real workflow, and map the real switching cost of an agent tool before the business becomes dependent on it.

The short version

  • On August 17, 2026, Google's A2A protocol (agent-to-agent communication) joined the Agentic AI Foundation, the Linux Foundation body that already governs Anthropic's MCP (agent-to-tool access), putting the two leading open AI agent standards under one neutral governance structure.
  • This was a governance change, not a technical one. A2A's specification did not change, and existing implementations need no update.
  • AAIF has grown from 49 founding members in December 2025 to more than 250 today, with AWS, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft, and OpenAI among its platinum members.
  • Shared governance reduces the risk that one company unilaterally changes or abandons a protocol, but it does not guarantee that any specific vendor's product implements the standard completely or actually works with a competitor's tool.
  • "Open standard" and "MCP compatible" claims are becoming a real purchasing signal. Verify them by asking a vendor for a working example with an independent tool, not by taking the label at face value.
  • If a business already uses more than one AI agent tool, ask each vendor directly where they stand on MCP and A2A adoption, since that answer says more about real flexibility than marketing language does.

Where ATLACIS can help

Sources

Tags:AI vendorsvendor dependencyAI buying decisionsAI agentsAI standardsbusiness AIAI decision supportAI infrastructure
FAQ

Common questions

What is the difference between MCP and A2A?
MCP (Model Context Protocol) standardizes how an AI agent connects to outside tools and data, like a database or a document library. A2A (Agent2Agent) standardizes how one AI agent communicates and hands off work to another agent, potentially built by a different company. They solve different problems and now share the same neutral governance body.
Does this mean every AI tool now works with every other AI tool?
No. It means the two standards that make that kind of interoperability possible now sit under one neutral foundation instead of being scattered across separate homes. Whether a specific vendor's product actually implements either standard completely, and whether it has been tested working with a competitor's tool, is still something to verify directly with that vendor.
Should I choose an AI vendor because they say they support open standards?
Treat it as a starting point, not a guarantee. Ask which version of the standard they implement, whether they have a working example with an independent tool, and whether they add proprietary extensions that would make switching away from them harder than the open standard implies.
Keep reading

More from the blog

Stripe just bought the tool built to prevent AI vendor lock-in. Here is what business owners should know before trusting a 'neutral' AI gateway.

Bloomberg reported on August 16, 2026 that Stripe has agreed to acquire OpenRouter, the AI model gateway millions of developers use to avoid depending on a single AI vendor, for more than $7 billion. The tool built to prevent lock-in is now owned by one company. Here is what that means for any business relying on a gateway, router, or other 'vendor agnostic' layer in its AI stack.

The AI vendor default era is over. Here is what business owners should do instead.

For three years, most businesses defaulted to OpenAI. In May 2026, Anthropic overtook OpenAI in US business AI spending for the first time, according to the Ramp AI Index. DeepSeek went from near zero to the fastest-growing measured vendor in the same period. The AI vendor market has fragmented, switching costs are near zero, and the decision is no longer about picking the right tool. It is about building so you can change your mind.

IBM surveyed 1,000 executives on AI vendor risk. 91 percent do not know what they depend on. Here is what that means for your business.

A June 2026 IBM Institute for Business Value study found that 91 percent of senior executives do not fully understand their AI dependencies. The research puts a profit number on that gap. The lesson is not to buy more AI. It is to understand what you already depend on.

Make better AI decisions, starting with one call.

Book a free AI Fit Call. We will tell you what to use, what to avoid, and where to start. No jargon, no pressure.

Book a CallSee How It Works