Google Announces A2A: An Open Protocol for AI Agent Collaboration

Google Launches Open A2A Protocol for AI Agent Collaboration

Enterprises are increasingly deploying autonomous AI agents to automate tasks, enhance productivity, and streamline complex workflows. However, the true potential of these agents is often limited by their inability to communicate and collaborate across different systems and platforms. To address this challenge, Google, along with over 50 technology and service partners, has announced the Agent2Agent (A2A) protocol.

A2A is a new, open protocol designed to enable AI agents, regardless of their underlying framework or vendor, to communicate securely, exchange information, and coordinate actions. This initiative aims to foster a dynamic, multi-agent ecosystem, breaking down silos between data systems and applications.

Partners supporting A2A include major technology players like Atlassian, Box, Cohere, Intuit, Langchain, MongoDB, PayPal, Salesforce, SAP, ServiceNow, UKG, and Workday, as well as leading service providers such as Accenture, Deloitte, KPMG, PwC, and TCS. This broad collaboration underscores a shared vision for a future where AI agents work together seamlessly.

Core Design Principles

The A2A protocol was developed based on five key principles:

• Embrace agentic capabilities: Focus on enabling natural, unstructured collaboration between agents, even without shared memory or tools.
• Build on existing standards: Leverage familiar standards like HTTP, SSE, and JSON-RPC for easier integration with existing IT infrastructure.
• Secure by default: Incorporate enterprise-grade authentication and authorization, aligning with OpenAPI's schemes.
• Support for long-running tasks: Designed to handle tasks ranging from quick queries to complex research that might take hours or days, providing real-time feedback and status updates.
• Modality agnostic: Support various data types beyond text, including audio and video streaming.

How A2A Facilitates Interaction

A2A defines interactions between a "client" agent (initiating tasks) and a "remote" agent (acting on tasks). Key capabilities include:

• Capability Discovery: Agents advertise their functions via a JSON-based "Agent Card," allowing client agents to find the right remote agent for a task.
• Task Management: Communication revolves around a defined "task" object with a lifecycle. Tasks can be completed quickly or span longer durations, with agents syncing on status. The output is termed an "artifact."
• Collaboration: Agents exchange messages containing context, replies, artifacts, or user instructions.
• User Experience Negotiation: Messages contain "parts" (e.g., text, images) with specified content types, enabling agents to negotiate formats and UI capabilities (like iframes or forms).

Example: Streamlining Candidate Sourcing

Imagine a hiring manager using an agent within a unified interface to find software engineer candidates based on specific criteria. This primary agent could use A2A to interact with specialized agents:

1. One agent searches internal HR systems.
2. Another agent scrapes public job boards or professional networks.
3. Results are consolidated and presented to the manager.
4. The manager directs their agent to schedule interviews via another A2A-enabled scheduling agent.
5. Post-interview, a background check agent is engaged.

This demonstrates how A2A enables agents across different systems to collaborate on a complex workflow.

The Future is Collaborative and Open

Google and its partners believe A2A will unlock significant value by enabling agents to work across entire enterprise application landscapes. By releasing the protocol as open source, they aim to build it collaboratively with the community.

Partner feedback highlights the enthusiasm for a standardized approach:

• Atlassian: "A standardized protocol like A2A will help agents successfully discover, coordinate, and reason with one another..."
• Cohere: "...the open A2A protocol ensures seamless, trusted collaboration..."
• SAP: "...a pivotal step toward enabling SAP Joule and other AI agents to seamlessly work across enterprise platforms..."
• ServiceNow: "ServiceNow and Google Cloud are collaborating to set a new industry standard for agent-to-agent interoperability..."

The draft specification, code samples, and contribution guidelines are available on the A2A website, with a production-ready version planned for later this year.

Source: Google Developers Blog - Announcing the Agent2Agent Protocol (A2A) by Rao Surapaneni, Miku Jha, Michael Vakoc, Todd Segal