Fahad Manzur
Google has moved its Agent2Agent protocol, the open standard that lets AI agents from different vendors find each other and hand off work, under the Agentic AI Foundation. The foundation, which already hosts Anthropic Model Context Protocol, said A2A now has backing from more than 150 organisations and is running in production across supply chains, financial services and mobile platforms. The change puts the two main agent interoperability standards under the same neutral governance.
The problem A2A was built to remove
Before a shared protocol existed, agents built on different frameworks could not pass work to each other without custom integration code written for that specific pairing. Every new vendor relationship meant repeating the same plumbing work. As the Agentic AI Foundation put it, the cost was never in the agents themselves, it was in the integrations between them.
A2A handles that with something called an agent card: a structured description of what an agent can do and how to reach it. Other agents read the card, discover the capability and delegate the task without a person brokering the handoff. Version 1.0, released in March, added multitenancy, version negotiation, multi-protocol bindings and signed agent cards for cryptographic identity verification.
Where it is already running
Adoption is broader than most people realise. Huawei has standardised on A2A to connect Celia, its operating-system-level assistant, with in-app agents across HarmonyOS. Tencent WeChat uses it to integrate with Huawei and other Android assistants. Google Cloud, Microsoft Azure AI Foundry and AWS Bedrock AgentCore all support hosting or discovering A2A agents, so a handoff can cross cloud providers. Google Cloud and PayPal are extending the idea into payments through a related protocol.
Google originally launched A2A in April 2025 and donated it to the Linux Foundation that June, alongside AWS, Cisco, Microsoft, Salesforce, SAP and ServiceNow. IBM Agent Communication Protocol was merged into A2A in August 2025.
The objection nobody has answered yet
Not everyone thinks consolidation solves the hard part. Mahesh Shanmugasundaram, lead AI solutions architect at Seekr, told Techstrong.ai that interoperability widens an existing trust gap rather than closing it. His concern is what he called an AI game of telephone: in an A2A chain, each agent tends to treat the previous agent output as fully trusted input instead of a claim to check. A small hallucination early in the chain can surface several steps later looking authoritative.
That distinction is worth holding onto. A protocol standardises the pipe. It says nothing about whether what travels through the pipe is accurate or appropriate for the action it triggers.
Reading this as someone who builds for clients
For developers and agencies taking on automation work, the practical signal is that the integration layer is settling. Two protocols now sit under one foundation, three major clouds support them, and that reduces the odds of building on a standard that disappears in eighteen months. Learning A2A concepts, agent cards, capability discovery, signed identity, is starting to look like a durable skill rather than a bet.
The second signal is commercial. If agent handoffs become cheap, the billable work shifts from writing connectors to designing verification: deciding what evidence an agent must attach to its output before another agent acts on it. Clients will pay for that once a chain fails on them publicly. This connects directly to the wider question of whether people will actually use AI agents at scale.
