Google Unveils A2A Protocol for AI Agents to Collaborate

Google has unveiled Agent2Agent (A2A), an open protocol designed to let AI agents built by different vendors communicate with one another, exchange information securely, and coordinate actions. The announcement arrives with backing from more than fifty technology partners and over a dozen consulting firms, giving the launch a weight that's unusual for a brand-new specification.

The idea behind it is easy to state and hard to solve: companies today are deploying agents that automate specific tasks — from ordering a new laptop to helping a customer service rep or assisting with supply chain planning — but those agents are trapped inside systems and applications that don't talk to each other. A2A aims to give them a common language.

What A2A Is and What Problem It Solves

According to Google's announcement, A2A is an open protocol that lets AI agents communicate, share information securely, and coordinate actions across different platforms or enterprise applications. The key detail is that this communication works even when the agents were built by different vendors or on different development frameworks.

That detail is the whole point of the proposal. In the current state of agentic AI — the approach in which a model doesn't just respond but autonomously carries out tasks — each company tends to build its agents within its own ecosystem. The result is a set of silos: an agent from one project management tool has no standard way to ask something of a billing or HR agent built by a different vendor.

Google argues that letting these agents interoperate "will increase autonomy and multiply productivity gains, while lowering long-term costs." The company says it designed the protocol by drawing on its own experience scaling agent systems, and to address the problems it identified while deploying large-scale multi-agent systems for its customers.

More Than 50 Partners on Board

The sheer number and profile of the companies backing the launch is the most striking part of the story. The technology partners include Atlassian, Box, Cohere, Intuit, Langchain, MongoDB, PayPal, Salesforce, SAP, ServiceNow, UKG and Workday — a list spanning enterprise software giants, payment companies and database providers.

Joining them are major service providers and consultancies: Accenture, BCG, Capgemini, Cognizant, Deloitte, HCLTech, Infosys, KPMG, McKinsey, PwC, TCS and Wipro. The fact that the consultancies that implement software at large corporations are on board from day one is no small detail — in practice, they're the ones who end up deciding which standards actually get deployed inside enterprises.

The presence of companies whose core business is management software — Salesforce, SAP, ServiceNow, Workday — points to exactly where Google wants A2A to take root: complex enterprise workflows, the kind that span multiple departments and multiple applications at once.

How It Fits with MCP, Anthropic's Standard

One point Google is careful to stress is that A2A doesn't compete with Anthropic's Model Context Protocol (MCP) — it complements it. That distinction matters for understanding what each piece actually does.

MCP, as the announcement itself describes it, gives agents tools and context: it's the mechanism through which an agent accesses the data, documents or functions it needs to do its job. A2A operates on a different layer — it's what lets one agent talk to another agent. One connects a model to its tools; the other connects agents to each other.

That division of labor sketches out what's starting to look like a stack of standards for agentic AI, with layers solving different problems. The fact that two of the industry's biggest players — Google and Anthropic — are pushing protocols presented as compatible rather than rival is a signal that, for now at least, the market would rather carve up the territory than fragment it into incompatible formats.

Why It Matters for Businesses

For an organization already using several tools with agent capabilities, A2A's promise is concrete: the ability to combine agents from different vendors and manage them in a standardized way across diverse platforms and cloud environments. Google sums it up by saying its customers' agents "will now be able to work across their entire enterprise application estates."

In practical terms, that points toward avoiding vendor lock-in. If one company's agent can hand off part of a task to another vendor's agent without custom-built integrations, a business gains the freedom to pick the best tool for each function instead of being tied to a single ecosystem.

Still, it's worth reading the announcement with the caution a specification deserves in its early days. An open protocol backed by fifty-odd names is a starting point, not a guarantee of adoption. The history of tech standards is littered with promising proposals that stalled halfway because the real implementation work — security, governance, cross-version compatibility — turned out to be far costlier than the initial paper support suggested.

The open question is whether A2A becomes the common language of enterprise agents or just another layer coexisting with alternatives. What the move does make clear is that interoperability, not just each model's raw power, has become a central battleground in agentic AI. The coming months will show how many of those fifty partners move from announcement to actual integration.