Google Unveils Agent Development Kit and A2A Protocol for AI Agent Connectivity

Google has introduced a new open-source development kit along with a protocol aimed at enhancing the communication capabilities of artificial intelligence (AI) agents. The initiative, known as Agent2Agent (A2A), has been developed in collaboration with 50 industry partners for Google Cloud's Vertex AI platform. This new protocol is designed to facilitate agent communication, enabling AI agents to express their requirements to one another more effectively.

The newly launched Agent Development Kit (ADK) is currently tailored for Python, with plans to extend support to additional programming languages in the future. Google Cloud asserts that developers can create an AI agent with fewer than 100 lines of code. Key features of the ADK include the ability to configure the reasoning processes of agents and define the systems they can interact with, incorporating specific guardrails to prevent unauthorized actions and protect sensitive data from leaks.

Interactions with these agents can occur through text, audio, or video formats. Within the Vertex AI platform, over 130 foundational models are available, including advanced models such as Gemini 1.5 Pro, totaling more than 200 models provided by various contributors, including Mistral, Meta, and Anthropic, among others. In addition to A2A, data can also be transmitted securely using the Model Context Protocol (MCP), originally developed by Anthropic.

The deployment of these agents can take place within Vertex AI or on Kubernetes, allowing for direct integration into operational environments. To ensure brand compliance in corporate contexts, mechanisms such as content filters, defined output limits, and prohibited topic areas will be implemented. Given that agents can assume user identities, a dedicated identity management system with associated permissions will be established to monitor behaviors in real-time, although specific details on this monitoring have yet to be disclosed.

The concept of software development kits for agents is not entirely new, as OpenAI previously released its own Agents SDK for GPT models, which can also be utilized for open-source models. Similarly, Amazon has developed its Bedrock Agents, which are undergoing continuous improvements. With A2A, Google aims to standardize inter-agent communication, allowing for compatibility with MCP. This interoperability will facilitate the collaboration between a client agent, which understands user needs, and a remote agent, which executes tasks.

Notable partners supporting the A2A initiative include Box, Intuit, Cohere, Atlassian, MongoDB, Salesforce, ServiceNow, PayPal, and SAP. The implementation will also involve major consulting firms such as McKinsey, BCG, KPMG, PwC, Wipro, and Accenture, who are expected to expedite agent-based process optimizations for end users. Google Cloud believes that the A2A framework will significantly benefit customers by enabling their AI agents to work seamlessly with existing enterprise applications.

For collaborative AI agents to reach their full potential, universal interoperability is essential. A2A employs established protocols like SSE, JSON-RPC, and HTTP for authorization and authentication, matching the capabilities offered by competitors like OpenAI. With A2A and the ADK, Google envisions the creation of genuine multi-agent scenarios, transforming agents from mere tools into autonomous entities capable of completing both quick tasks and extensive projects, such as deep research requiring hours or even days of processing time, necessitating human oversight at critical points.

Real-time feedback is incorporated through a dedicated notification protocol. While Google has not yet provided pricing details regarding the integration of A2A and ADK into the Vertex AI framework, a draft specification and example code are available on GitHub. Further information and a production-ready version of A2A are anticipated in the upcoming months, with Google Cloud relying on its partners for implementation. The company is optimistic that AI agents will enhance productivity by autonomously handling numerous repetitive or complex daily tasks.