One of the significant announcements at the Google Cloud Next event in the early part of last month was the introduction of Protocol Agent2Agent, designed to enable AI agents to converse with each other.
The Protocol Agent2Agent (A2A) is Google’s endeavor to establish standards for the era of AI agents, following Google’s lead in the generative AI era. Examples include the API of the LLM model, which should adhere to OpenAI’s API or protocols that connect the LLM model with external data sources using Anthropic’s MCP.
Google explicitly explains that A2A is designed to complement MCP because they function differently (see accompanying diagram). Where MCP connects models to data sources or external tools & resources, A2A connects agents to engage in agent-agent collaboration.
An example of using A2A showcased by Google is directing an AI agent to find suitable job applicants based on factors like job position, location, and skills. Once the first agent filters the candidates, the second agent sets up interview appointments, and the third agent verifies applicants’ backgrounds. Having multiple agents working together reduces the workload, and these agents can exchange commands and data through the A2A protocol.
A visual demonstration of utilizing multiple agents through Google’s Agentspace.
The overview of A2A involves client agents issuing commands and remote agents executing tasks, with concepts of capability discovery, task management, and negotiation to determine outcomes such as creating images, videos, or filling out forms.
Negotiating to display applicant information as plain text or embedding it attractively in an iframe is one example of negotiations.
The A2A protocol operates on existing open standards, like sending metadata information on what each agent can do as JSON and the server agent being an HTTP endpoint.
The intriguing aspect of A2A this time is Google’s readiness and strong support from various software companies like Atlassian, JetBrains, SAP, Oracle, MongoDB, Salesforce, ServiceNow, Elastic, Datastax, Workday, as well as IT consulting firms like Accenture, BCG, Deloitte, Infosys, KPMG, McKinsey, PWC, Wipro.
Google is also preparing tools to integrate A2A with various agent frameworks in the market, such as LangGraph, Genkit, LlamaIndex, CrewAI, Semantic Kernel, Marvin, along with its Agent Development Kit (ADK). It remains to be seen if Google’s preparations for A2A will meet the expectations set by Google.
Source: Agent2Agent, Google Developers
TLDR: Google introduced Protocol Agent2Agent at Google Cloud Next event, enabling AI agents to converse. A2A aims to set standards for the AI agent era, fostering collaboration among agents for tasks like candidate screening and background verification. Support from various software and IT consulting companies indicates widespread industry readiness, with Google preparing tools to integrate A2A with multiple agent frameworks. Expectations are high as Google moves forward with A2A implementation.
Leave a Comment