Google teams up with 50+ partners to launch A2A agent protocol

Every enterprise has started building AI agents to handle routine tasks, from ordering laptops to managing customer tickets. But there's a glaring catch: your Workday agent doesn’t know your Salesforce agent exists. Whenever a process crosses software borders, a human still has to play messenger.

Google introduced Agent2Agent (A2A), an open protocol designed to let AI agents securely communicate, share information, and coordinate actions across different frameworks and platforms.

An open language for AI agents

Google didn't launch this in a vacuum. It assembled a coalition of more than 50 enterprise heavyweights and tech partners, including Atlassian, Box, Cohere, Langchain, MongoDB, PayPal, Salesforce, SAP, ServiceNow, and Workday, along with top consultancies like McKinsey, BCG, and Deloitte.

Under the hood, A2A relies on a client-and-remote relationship. A client agent translates a user's goal into specific subtasks, finds the right remote agent, and delegates execution.

Take software engineering hiring as an example. Instead of jumping between tabs, a hiring manager can simply instruct their agent to find matching candidates. That agent pings a specialized recruiting agent to source profiles, passes results back to the user, triggers a scheduling agent to book interviews, and brings in another agent for background checks once offers are made.

Google explicitly positioned A2A as a complement to Anthropic’s Model Context Protocol (MCP). While MCP standardizes how individual agents connect to local tools and data context, A2A handles horizontal agent-to-agent negotiation across disparate platforms.

Google released the draft specification and code samples as open source, with plans to launch a production-ready version with its partners later this year.

Why it matters

We think A2A addresses the single biggest bottleneck threatening enterprise agent adoption: multi-agent fragmentation. Without an open standard, every SaaS provider would trap customers inside their own walled gardens, forcing developers to build brittle point-to-point API connectors for every agent interaction.

For enterprise teams, A2A means you won't have to rewrite your agent stack just because your HR tools and customer databases live on different clouds. But because the specification is currently in draft form, it is not something businesses can deploy at scale tomorrow. For now, engineering teams should evaluate the specs to design future-proof agent architectures rather than rushing into production.

Strategically, Google played this smartly by rallying enterprise competitors around an open protocol rather than trying to own the entire agent runtime. Still, real-world execution will be tough. Getting 50 vendors to agree on authentication, token management, error handling, and audit trails when autonomous agents operate across company firewalls will be much harder than releasing a specification.