Google LLC today launched a new,open protocol called Agent2Agent that allows artificial intelligence agents to interoperate securely with each other even if they’re built by different developers or on a separate framework.

Recently there has been an explosion of new autonomous AI agents and frameworks built within every industry helping businesses and employees to handle recurring and complex tasks that focus on specific tasks. As a result, the technology is becoming increasingly fragmented. Allowing agents to call upon each other to broaden their capabilities, instead of isolating them within the boundaries of business domains, can help increase their autonomy.

AI agents are part of an ongoing trend known as agentic AI, which refers to AI systems that can solve multi-step problems based on specific goals, going beyond question-answering or content generation. They can act autonomously and with little or no user supervision. AI agents are built with multiple large language models using complex reasoning and often must interact with external third-party tools.

As part of today’s launch, Google announced that the Agent2Agent protocol has the support of more than 50 technology partners including Atlassian, Box, Cohere, Intuit, Langchain, MongoDB, Salesforce, SAP, ServiceNow, UKG and Workday. Leading service providers such as Accenture, BCG, Capgemini, Cognizant, Deloitte and PwC also joined their ranks.

The A2A protocol works by providing a communication channel between one agent and another to allow the first to ask the other to complete a task. That means agents need to be able to advertise what they can do so that consumer agents can tell what other agents are capable of before connecting to them. Then they can negotiate what tasks need to be completed and collaborate on context, replies, artifacts – such as files, video, documents and other content — and other user instructions.

“With Atlassian’s investment in Rovo agents, the development of a standardized protocol like A2A will help agents successfully discover, coordinate and reason with one another to enable richer forms of delegation and collaboration at scale,” said Brendan Haire, vice president for engineering of AI Platform at Atlassian.

Agent collaboration enabled by A2A will allow collaboration across numerous business processes that would otherwise have to stop and have users take it to the next step using a different tool, Google said.

For example, in a hiring process, a manager could ask an agent to discover potential candidates using different job listings, locations and skill sets. The agent could then reach out to different agentic AI systems at staffing firms to source potential candidates. The user could then read through the suggestions, using the information it surfaced to get a handle on who to contact and direct the agent to schedule interviews. After the interviews are complete, another agent can assist with doing background checks, which could require reaching out to another agent platform.

That would greatly reduce the number of interactions and touchpoints that a user would have to make, by allowing an AI agent to do most of the busywork behind the scenes by researching candidates, scheduling interviews, contacting staffing firms and doing background checks. All of these are multistep processes that require advanced reasoning to determine the best way to complete the task and report back.

A2A complements a toolset that connects LLMs to data in external systems called Model Context Protocol, or MCP for short, developed by Anthropic PBC. MCP provides helpful tools for agents so that developers do not need to write numerous lines of custom code to allow LLMs to talk to external databases or other systems. According to Google, MCP will enable agents to connect easily with a broader variety of business systems, databases and cloud environments.

Google has published the full specifications on GitHub, including code samples and further example scenarios for open-source and business developers to understand how they can contribute to the protocol ecosystem. The company said it’s working with partners to launch a production-ready version of the protocol later this year.

Image: SiliconANGLE/Microsoft Designer