Summary: Agent2Agent marked one year with more than 150 supporting organizations, up from more than 50 a year earlier. Its v1.0 specification shipped in March 2026, adding signed Agent Cards, multi-tenancy, and production migration support.
Agent2Agent — A2A for short — turned one on April 9, 2026, and it has numbers to show: the Linux Foundation, marking the date, counts 150-plus organizations behind the project, up from 50-plus a year earlier; a core repository past 22,000 GitHub stars; and an SDK collection that began with Python alone and now covers five languages. Started as a Google project and now shepherded by the Linux Foundation, this open standard is settling in as the go-to answer for getting agents from different organizations to interoperate.
The candles matter less to developers than the spec: v1.0 came out in March 2026 — the first release you can actually run in production — migration path in hand. That fight over standards is settling quicker than almost anyone predicted.
The First Year: 50 Organizations Become 150+
On April 9, 2025, Google took the wraps off A2A, so that an agent on one framework and an agent on another could share a single language. June 23, 2025 — two months after the announcement — brought the handover at Denver’s Open Source Summit North America: Google placed A2A under the Linux Foundation, and from that point no single vendor owned it — the community did. On the founding roster: ServiceNow, SAP, Salesforce, Microsoft, Cisco, and AWS.
Twelve months on, membership tops 150 — IBM among others signed on as the year went on — and the production work clusters in insurance, IT operations, supply chain, and financial services. By the Linux Foundation’s telling, this is the first open standard of its kind that is genuinely production-ready, built for agent-to-agent communication worldwide. For developers, the most tangible change is in the SDKs: they used to mean Python and nothing else, and now span five languages teams can run in production — Python, JavaScript, Java, Go, and .NET — so most groups can treat the protocol as a native capability instead of propping up the reference code on their own.
The Stable v1.0 Spec, with Signed Agent Cards Up Front
v1.0 wrapped up in March 2026 with support for multiple protocols, multi-tenancy built for enterprises, a refreshed set of security flows, plus a path for migration. Signed Agent Cards is the feature getting top billing: an agent’s identity gets checked through cryptographic signatures, so before a task goes to an agent run outside the delegating organization, that side can pin down who is really behind it.
Web conventions carry part of the design, too: the security and load-balancing setups companies already run plug straight in, and that is where its at-scale reliability comes from. On Google’s open source blog, demos walked through ADK, LangGraph, AG2, and CrewAI working together — an agent on one platform passing a sub-task to an agent on another, with nobody’s internal memory changing hands.
Where MCP Stops and A2A Starts, Plus AP2 for Payments
Here is the official one-line division of labor: MCP is about wiring agents into internal tooling and data; A2A is about coordination between agents whose owners sit in different organizations. One completes the other — they are not rivals.
Built on the open extensibility that A2A exposes, an “A2Family” is taking shape as well. AP2 — the Agent Payments Protocol — gives agents a safe route through a transaction, and over 60 organizations across payments and financial services stand behind it. UCP slots in via AP2’s mandates extension, one that keeps cryptographic proof a user consented to a purchase. A2UI, finally, is the layer where agents put interfaces in front of people.
Landing on the Big Clouds, and What Comes Next
Where deployment is concerned, Copilot Studio and Azure AI Foundry support A2A natively on Microsoft’s side, while AWS’s Amazon Bedrock AgentCore Runtime does the same on its platform. Google Cloud’s VP and GM, Rao Surapaneni, framed it plainly: how much an AI agent is worth tracks how well it can work with others. Cisco Distinguished Engineer Luca Muscariello sees A2A as “the syntactic layer that makes agent-to-agent communication reliable and interoperable.”
Still queued on the roadmap are four pieces of work — best practices for secure deployment, broader testing plus tooling, consolidating the registries, and a specification aimed at interoperability. Anyone who wants to learn can pick up DeepLearning.AI’s short A2A course; samples and tutorials from the official project are available too.
What This Means for Developers
Two conclusions. One: for agents that need to interoperate, the standards question is basically answered — with the Linux Foundation holding governance in neutral, pairing A2A with MCP is where the major frameworks and the big clouds have all landed, so a new project can treat that pairing as given rather than rolling a wire protocol of its own, and no choice of framework walls you off from workflows that span organizations. Two: payment and identity — the thorniest trust problems in agent workflows that span organizations — are each getting a standard: AP2 for the first, Signed Agent Cards for the second. Building blocks like these are what enterprise AI’s next phase stands on, and they showed up before teams would otherwise have botched those problems on their own.