Something unusual happened in the AI industry this month: the giants agreed on something. Google's Agent2Agent protocol, known as A2A, formally became a hosted project of the Agentic AI Foundation, the Linux Foundation body created in December 2025 to steward the open infrastructure of AI agents. That puts A2A under the same neutral roof as Anthropic's Model Context Protocol (MCP), Block's Goose framework and OpenAI's AGENTS.md convention, and the foundation's membership has grown from fewer than forty companies at launch to more than 250, including Amazon, Anthropic, Bloomberg, Google, Microsoft, OpenAI and Shopify.
The two protocols solve complementary problems, and the division of labour is easy to picture. MCP is how an agent talks to tools and data: your calendar, your database, your document store. A2A is how an agent talks to another agent: a purchasing agent negotiating with a supplier's sales agent, a support agent handing a case to a logistics agent. One connects agents downward to systems; the other connects them sideways to each other. With both now governed by one open foundation backed by every major laboratory, the plumbing of the agent economy is standardizing the way the web's plumbing did in the nineties, before the interesting layers were built on top.
Why should a business that simply wants a working agent care about foundation governance? Because the alternative was a standards war, and in standards wars the casualties are the customers. Every company that bought a proprietary agent stack in the last two years faces the same questions: can this agent use tools from another vendor, can it cooperate with an agent my supplier runs, and what survives if I switch providers? Open protocols under neutral governance convert those questions from bets into checkboxes. History is on the customer's side here: nobody asks which browser your website works with anymore.
The practical takeaway fits in three vendor questions. Does your agent expose and consume open protocols, MCP for tools and A2A for agent-to-agent work, or does it only speak its own dialect? What is the migration path for the knowledge and integrations we build with you? And which of the foundation's standards do you commit to tracking as they evolve? A vendor comfortable with those questions is selling you an asset; a vendor who changes the subject is selling you a dependency.
For our part, the answer is on the record: the agents we build run on Anthropic's stack, whose Model Context Protocol is one of the foundation's founding projects, and every deployment leaves the client with documented integrations and a handover manual precisely so the investment outlives any single vendor, ourselves included. Standards make that promise cheaper to keep, which is why we welcome this consolidation with no hesitation at all.