Google's NEW Agent2Agent Protocol

Summary

In this video, the speaker shares insights from day two of Google Cloud Next 2025, focusing on the developer keynote where the main theme revolved around AI agents. The speaker discusses the 'Agent2Agent' protocol announced by Google, which allows agents to communicate and collaborate directly, creating a more sophisticated version of agent-based workflows. Despite the buzz around agents, the speaker notes that many demos were underwhelming, showcasing basic workflows rather than fully agentic solutions. The video highlights key concepts of agent discovery and the potential future of agent marketplaces, but raises concerns about the absence of key industry players like Anthropic.

Takeaways

• 😀 The focus of day two at Google Cloud Next 2025 was on the developer keynote, where the word 'agents' was used extensively, highlighting their growing importance in tech.
• 😀 Despite the excitement around agents, the demos at the expo were underwhelming, showcasing basic workflows like currency converters rather than fully autonomous agents.
• 😀 The real takeaway was that we are still far from achieving true **Artificial General Intelligence (AGI)** with agents as the technology seen was more akin to LLM-based workflows than fully autonomous systems.
• 😀 Google introduced the **Agent2Agent (A2A) protocol**, which facilitates communication between agents, allowing them to collaborate in a more advanced way compared to previous models.
• 😀 A2A enables agents to discover and negotiate tasks with other agents, improving collaboration and task execution by sharing specifications like formats and languages.
• 😀 The A2A protocol uses established standards such as **HTTP**, **JSON-RPC**, and **SSE** for secure communication and data exchange between agents.
• 😀 Google is not replacing existing tools like **MCP** with A2A but is introducing it as a complementary protocol for inter-agent communication.
• 😀 One exciting aspect of the A2A protocol is the potential for an **agent marketplace**, where agents can 'shop' for others that best suit their tasks, creating a new ecosystem for agent-based collaboration.
• 😀 The absence of **Anthropic** from the list of partners supporting A2A raised concerns about possible fragmentation and the risk of conflicting protocols emerging within the agent landscape.
• 😀 Despite some concerns about adoption, Google is actively seeking feedback on the A2A protocol and its potential, emphasizing that it is open to community input and suggestions for improvement.
• 😀 Google’s **Agent Developer Kit** and sample agents show the practical potential of A2A, providing developers with the tools to experiment and build more sophisticated agent-based solutions.

Q & A

What was the main focus of the developer keynote at Google Cloud Next 2025?

The main focus of the developer keynote was on 'agents,' with the term being mentioned more frequently than even AI itself. The keynote largely revolved around discussing the future potential of agents and how they could transform workflows.

What were the demo agents presented at the conference like?

The demo agents presented at the conference were mostly small workflows or applications, such as currency converters. These were seen as less impressive and sophisticated compared to the fully agentic solutions people were hoping for.

What does the term 'Agent2Agent' (A2A) protocol refer to?

The Agent2Agent (A2A) protocol is a new system that allows agents to communicate and collaborate directly with each other. Unlike existing protocols that focus on agent-to-tool communication, A2A enables agents to discover and negotiate tasks between them.

What are the key features of the Agent2Agent protocol?

The A2A protocol allows agents to communicate securely with one another, using standards like HTTP and JSON-RPC. It enables agent discovery, where one agent can find others to collaborate with, and facilitates task negotiation, ensuring that agents can specify how tasks should be handled.

How does Google plan to ensure security in the A2A protocol?

Google has emphasized that the A2A protocol is designed to be secure by default. While they humorously mentioned that the 'S' in MCP doesn't stand for security, they are aiming to make the A2A protocol more secure than existing solutions.

What is the 'agent store' concept mentioned in the script?

The 'agent store' concept refers to the idea of a marketplace where agents can browse and select other agents to help them complete tasks. Agents would pay for the services of other agents, potentially based on the duration or complexity of the tasks.

What is the concern regarding the absence of Anthropic in the A2A protocol ecosystem?

The absence of Anthropic from the list of partners for the A2A protocol raises concerns about potential fragmentation in the agent ecosystem. If Anthropic develops its own version of an Agent2Agent protocol or tool communication, it could lead to competing protocols that hinder collaboration.

What role do 'agent cards' play in the A2A protocol?

Agent cards are key to the A2A protocol as they allow agents to advertise their capabilities to other agents. These cards contain information about what the agent can do, how it does it, and other relevant details, facilitating discovery and collaboration between agents.

What was the overall reaction to the demos seen at the conference?

The overall reaction to the demos was underwhelming. While the agents showcased were functional, they were seen as basic workflows rather than transformative, highly agentic solutions that could change industries or disrupt the workforce as some had hoped.

Why is the A2A protocol seen as a potential breakthrough for agent collaboration?

The A2A protocol is seen as a potential breakthrough because it enables direct communication and collaboration between agents, which opens up new possibilities for complex, multi-agent workflows. This could lead to more advanced agent ecosystems where agents can work together to solve tasks autonomously.