Google's Agent-to-Agent (A2A) protocol, a new open standard for AI agent interoperability, is gaining attention now that companies have invested in a variety of AI agents. The A2A protocol provides a common communication layer for discovery, secure interaction, and task delegation between Elastic AI agents.
In a recent DBTA webinar, Collaborative AI: A Practical Guide to Google's A2A Protocol and Elastic AI Agents, Travis Norenberg, partner solutions architect, GenAI, Google Cloud and Joe McElroy, principal software engineer, Elastic, discussed how A2A enables complex, multi-agent workflows.
Right now, agents face a variety of challenges, Norenberg and McElroy explained, including a fragmented landscape, it’s hard to integrate, and there’s a lack of ops and governance. How do we achieve faster development and iteration? How do we manage dependencies, delegate effectively, and avoid creating hard-to-debug workflows between agents? Beyond development, how do we reliably deploy, scale, monitor, and govern an ecosystem of agents?
The A2A protocol facilitates dynamic communication with different agents as peers, Norenberg and McElroy said. A2A capabilities must follow specific guidelines that include:
- Agents must advertise their capabilities, so clients know when and how to utilize them for specific tasks.
- Clients and agents need to agree on communication methods like text, forms, iframe, or audio/video to ensure proper user interaction.
- Clients and agents need mechanisms to communicate task status, changes, and dependencies throughout task execution.
- Clients and agents must support dynamic interaction, enabling agents to request clarifications, information, or sub-actions from clients, other agents, or users.
According to Norenberg, a possible agentic stack can include:
Agent Development Kit: Open-source, code-first toolkit for building, evaluating, and deploying AI agents.
Model Context Protocol: Open protocol that standardizes how applications provide context to LLMs. It’s complimentary to the A2A protocol.
Vertex AI Agent Engine: Managed platform to deploy, manage, and scale AI agents in production.
Agent2Agent (A2A) protocol: Open standard designed to enable communication and collaboration between AI agents.
With Google Cloud, users can build agents at their preferred level of abstraction. All code (Level 3 and below) can be deployed and autoscaled with Vertex AI Agent Engine.
ADK is a programmatic framework for developers to define, test, and iterate on multi-agent applications. ADK is open source.
The Agent Development Kit offers a framework and SDK to build multi-agent solutions. This system includes:
- Define multi-agent applications orchestrating actions across many agents and tools
- Intuitive local dev UI for fast iteration; visualize agent topology and trace agent’s actions
- Interleave deterministic logic with actions driven by gen AI reasoning for effortless, hybrid agents
- Built-in support for long-running sync tools / human-in-the-loop
- Gemini is the default, but any generative AI model is supported, including fine-tuned models
- Support to call remote agents and thousands of existing tools via MCP and A2A protocols
For the full webinar, featuring a more in-depth discussion, Q&A, demo, and more, you can view an archived version of the webinar here.