> ## Content Index > Fetch the complete content index at: https://www.edgewisely.com/llms.txt > Use this file to discover other available public pages before exploring further. # Google Gives Away the Handshake - URL: https://www.edgewisely.com/google-gives-away-the-handshake/ - Published: 2026-08-25T01:54:09.000Z - Updated: 2026-08-25T01:54:09.000Z - Description: How moving the A2A protocol to the Agentic AI Foundation trades control for the one thing enterprise buyers actually demand. - Author: John Karpentar - Tags: AI, Enterprise **In a market with no dominant platform, owning the standard is the reason nobody adopts it.** Google built a protocol for AI agents to talk to each other, spent a year signing up partners, watched it land in every major cloud — and then handed it to a foundation it does not control. On August 17, the [Agent2Agent protocol moved into the Agentic AI Foundation](https://www.axios.com/2026/08/17/a2a-agentic-ai-foundation-open-ai-standards?ref=edgewisely.com), the Linux Foundation body created in December 2025 to house the plumbing of agentic AI. A2A now sits alongside Anthropic's Model Context Protocol under the same neutral governance, in a foundation that has gone from fewer than 40 members to more than 250 in eight months. That is the news. The strategy underneath it is more interesting: in a market where no one can win the standards fight outright, the next best outcome is to make sure nobody else does either. ## What A2A does, and why it needed a home The two protocols solve adjacent problems and people confuse them constantly, so it is worth being precise. **MCP connects an agent to things that are not agents** — a database, a filesystem, a ticketing system, an internal API. It is the standard for how a model reaches tools and data. **A2A connects an agent to another agent**, potentially one built by a different company, on a different model, running inside a different organization entirely. It defines discovery, task handoff, and coordination across boundaries. You need both, and until this month they lived in different places. MCP was one of the [three founding projects of the Agentic AI Foundation](https://www.linuxfoundation.org/press/linux-foundation-announces-the-formation-of-the-agentic-ai-foundation?ref=edgewisely.com) alongside Block's goose agent framework and OpenAI's AGENTS.md convention. A2A was donated by Google to the broader Linux Foundation portfolio, adjacent but separate. The move consolidates them. The stated logic is exactly what you would expect. "Companies don't want just one protocol; they want the whole stack to be open," AAIF executive director Mazin Gilbert told Axios. Google Cloud VP Rao Surapaneni framed the original design assumption plainly: customers were already deploying agentic systems from multiple vendors, and those agents had to work together. Both statements are true. Neither explains the timing. ## The real problem is procurement, not engineering Here is what actually stalls agentic AI inside large enterprises, and it is not a technical limitation. A bank has a customer-service agent from one vendor, a fraud-detection agent built in-house on a second model, a document-processing agent that came with its ERP, and a coding agent its engineering team standardized on. Every pair of those systems that needs to cooperate is a bespoke integration — written by someone, owned by someone, and broken the moment either side ships a breaking change. The cost is not the connector. The cost is that the connector is a permanent liability on someone's roadmap, and that the vendor lock-in it creates is invisible until you try to leave. Procurement departments have learned to see this coming. The question that kills enterprise AI deals in 2026 is not "does it work?" It is "what happens when we want to swap you out?" A shared protocol under neutral governance is the answer to that question. It lets a buyer commit without committing — choose a provider on cost, latency, or capability today, and preserve the option to choose differently in eighteen months. Axios framed the payoff correctly: common protocols push the industry toward modular, model-agnostic systems, which gives companies leverage they do not currently have. That leverage is precisely what makes a single-vendor protocol unacceptable, no matter how good it is. It is the same instinct driving demand for [neutral routing layers between applications and models](https://edgewisely.com/stripes-7-5b-openrouter-bet/?ref=edgewisely.com) — buyers want the option to change their minds. ## Why Google gave it away Google's position on A2A was strong and structurally doomed. Strong, because the protocol worked and adoption was real. The Linux Foundation reported in the spring that A2A had [passed 150 supporting organizations](https://www.linuxfoundation.org/press/a2a-protocol-surpasses-150-organizations-lands-in-major-cloud-platforms-and-sees-enterprise-production-use-in-first-year?ref=edgewisely.com), landed in major cloud platforms, and reached production use inside enterprises within its first year. Those are not vanity metrics. Doomed, because a Google-governed protocol for cross-vendor agent communication asks Microsoft, Amazon, and OpenAI to route their agents' conversations through a specification that Google can change. As [Forbes noted in its analysis of the move](https://www.forbes.com/sites/janakirammsv/2026/08/19/agent2agent-joins-the-agentic-ai-foundation-alongside-mcp/?ref=edgewisely.com), consolidating the two protocols under one roof is as much a signal about governance as about engineering. No enterprise architecture committee signs off on that, and no competitor evangelizes it. The protocol's entire value proposition is neutrality, and Google could not credibly supply neutrality about itself. So Google traded control for adoption — and the trade is a good one, because Google was never going to monetize the protocol directly. It monetizes Google Cloud, Gemini, and Vertex. A widely adopted A2A makes multi-vendor agent deployments practical, and multi-vendor deployments are exactly the environments where Google can compete on merit instead of being locked out by an incumbent. Anthropic made the same calculation with MCP and got the same result: MCP became the default not despite being given away but because it was. There is a version of this that is pure altruism and a version that is pure self-interest, and they produce identical behavior. That is what a well-designed standards commons looks like. ## Who this lands on **For enterprise buyers,** this is the most useful development of the month, and it is almost entirely about risk. Two protocols under one neutral roof is a materially stronger foundation for a five-year architecture than two protocols under two vendors. It does not eliminate switching costs — model behavior, prompt engineering, and evaluation suites remain sticky in ways protocols do not touch — but it removes the layer that was hardest to argue with in a board deck. **For the platform vendors,** the competition moves up the stack. If interoperability is table stakes, differentiation has to come from model quality, agent reliability, observability, cost, and the boring operational work of running agents in production without them doing something expensive at 3 a.m. That favors whoever is best at the unglamorous parts — including the [containment and permissioning layer](https://edgewisely.com/obsidians-bet-on-the-agent-leash/?ref=edgewisely.com) that decides what an agent is allowed to touch. It is a worse world for anyone whose moat was a proprietary connector. **For startups,** it is a genuine opening. Building an agent that has to integrate individually with every incumbent platform is a losing proposition for a company of twelve people. Building an agent that speaks A2A and MCP and is therefore reachable by anything is a plausible one — the same leverage smaller vendors have found by [removing configuration work rather than adding features](https://edgewisely.com/writer-cuts-the-prompt/?ref=edgewisely.com). Standards lower the cost of being small, which is why incumbents historically resisted them and why it is notable that these ones did not. **For AAIF,** growth is now the risk. Going from 40 members to 250 in eight months is a triumph and a governance problem in the same sentence. Standards bodies with many powerful members and competing incentives are how you get design-by-committee, glacial spec cycles, and vendor extensions that quietly fragment the thing the body exists to unify. The platinum roster — AWS, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft, OpenAI — reads as legitimacy today. It reads as a coordination problem the first time two of them want incompatible things. ## The unfinished part Protocols do not make agents interoperable. They make agents *addressable*, which is a smaller and more necessary thing. A2A can tell your agent how to hand a task to another company's agent. It cannot tell you whether that agent will do the task correctly, what it will cost, who is liable when it fails, or what it did with the data you passed it. Those questions — trust, pricing, authorization, audit — are the actual hard parts of cross-organizational agent work, and they sit above the protocol layer where the standards work has barely started. The web took a decade to go from HTTP to a functioning commercial internet, and the gap was filled with things HTTP did not specify: TLS, payment rails, identity, reputation, and law. Agent interoperability is at roughly the HTTP stage. Consolidating the protocols is what makes the next decade of work possible. It is not the work. ## The zoom-out The pattern worth keeping is that both of the protocols that matter in agentic AI were built by companies that then gave them away, and both became more valuable to their creators as a result. That is not generosity and it is not naivety. It is the recognition that in a market with several strong competitors and no dominant platform, the value of owning the standard is negative — because owning it is the reason nobody adopts it. The winning move is to establish the standard, seed it, and then visibly relinquish control at the moment when neutrality becomes worth more than authorship. Google appears to have timed that moment well. The harder judgment is whether AAIF can stay neutral once 250 members with real money at stake start filing competing proposals. Standards bodies rarely fail at launch. They fail at scale. *The instinct to own the protocol is usually the instinct that kills it.* ## Frequently Asked Questions ### What is the difference between A2A and MCP? MCP, the Model Context Protocol, connects AI agents to tools and data sources such as databases, APIs, and internal systems. A2A, the Agent2Agent protocol, handles communication between independent agents that may be built by different companies on different models. Enterprises deploying multi-vendor agent systems generally need both protocols working together. ### Why did Google hand A2A to the Agentic AI Foundation? A protocol governed by a single vendor is difficult for competitors to adopt, because it gives that vendor unilateral control over a shared dependency. Moving A2A to neutral governance under the Linux Foundation removes that objection. Google does not monetize the protocol directly, so wider adoption in multi-vendor environments is worth more than control. ### How large is the Agentic AI Foundation? AAIF launched in December 2025 with fewer than 40 members and has grown to more than 250 as of August 2026\. Platinum members include Amazon Web Services, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft, and OpenAI. Its founding projects were MCP, Block's goose agent framework, and OpenAI's AGENTS.md standard. ### Does A2A actually make AI agents interoperable? Only partially. A2A standardizes how agents discover each other and exchange tasks, which makes them addressable across organizational boundaries. It does not address trust, pricing, liability, authorization, or auditability — the questions that determine whether cross-company agent work is commercially viable. Those layers remain largely unbuilt. --- **Editor's note — sources:** - Axios — https://www.axios.com/2026/08/17/a2a-agentic-ai-foundation-open-ai-standards - Linux Foundation, AAIF formation — https://www.linuxfoundation.org/press/linux-foundation-announces-the-formation-of-the-agentic-ai-foundation - Linux Foundation, A2A milestone — https://www.linuxfoundation.org/press/a2a-protocol-surpasses-150-organizations-lands-in-major-cloud-platforms-and-sees-enterprise-production-use-in-first-year - Forbes — https://www.forbes.com/sites/janakirammsv/2026/08/19/agent2agent-joins-the-agentic-ai-foundation-alongside-mcp/