Google’s AI Agent Ecosystem Overview

Here is a short and crisp blog detailing the Agent Development Kit (ADK), Vertex AI Agent Engine, Agent2Agent Protocol (A2A), and Agentspace ( it covers details of my session for an AI Meetup in Bangalore (All things AI)

1. Agent Development Kit (ADK)

• What it is: A new open-source framework from Google, introduced at Google Cloud NEXT 2025 on April 9, 2025. It is the same framework that powers agents within Google products like Agentspace and the Google Customer Engagement Suite (CES).
• Purpose: Designed to simplify the full stack end-to-end development of agents and multi-agent systems, empowering developers to build production-ready agentic applications with greater flexibility and precise control. It is built to be flexible, use different models, and build production-ready agents for various deployment environments.
• Core Pillars / Capabilities across the agent development lifecycle:
• Multi-Agent by Design: Build modular, scalable applications by composing multiple specialized agents in a hierarchy, enabling complex coordination and delegation.
• Rich Model Ecosystem: Works with models like Gemini, any model via Vertex AI Model Garden, and offers LiteLLM integration for providers like Anthropic, Meta, Mistral AI, and AI21 Labs.
• Rich Tool Ecosystem: Equip agents with diverse capabilities using pre-built tools (Search, Code Exec), Model Context Protocol (MCP) tools, 3rd-party libraries (LangChain, LlamaIndex), or even other agents as tools (LangGraph, CrewAI).
• Built-in Streaming: Interact with agents in human-like conversations using bidirectional audio and video streaming capabilities, moving beyond text into rich, multimodal dialogue.
• Flexible Orchestration: Define workflows using workflow agents (Sequential, Parallel, Loop) for predictable pipelines, or leverage LLM-driven dynamic routing (LlmAgent transfer) for adaptive behavior.
• Integrated Developer Experience: Develop, test, and debug locally with a powerful CLI and a visual Web UI; inspect events, state, and agent execution step-by-step.
• Built-in Evaluation: Systematically assess agent performance by evaluating final response quality and step-by-step execution trajectory against predefined test cases.
• Easy Deployment: Containerize and deploy agents anywhere, including through integration with Vertex AI Agent Engine.
• Building Multi-Agent Systems: ADK excels in building collaborative multi-agent systems through hierarchical structures and intelligent routing. Delegation works by the LLM considering the query, the current agent’s description, and sub-agents’ descriptions to determine the best fit for a task, initiating a transfer if needed.
• Optimization: While flexible, ADK is optimized for seamless integration within the Google Cloud ecosystem, particularly with Gemini models and Vertex AI, offering a direct path to deploy agents onto Vertex AI’s managed runtime for scalability. It enables agents to connect to systems and data via over 100 pre-built connectors, Application Integration workflows, and access data in systems like AlloyDB, BigQuery, NetApp, and Apigee APIs.
• Comparison with Genkit: ADK is optimized for complex agents and multi-agent systems, providing higher-level abstractions and built-in integration for LiteLLM and Vertex AI Model Garden, along with support for bidirectional streaming. Genkit, conversely, provides fundamental building blocks for a wider variety of AI-powered experiences.

2. Vertex AI Agent Engine

• What it is: A set of services that enables developers to deploy, manage, and scale AI agents in production. It was formerly known as LangChain on Vertex AI or Vertex AI Reasoning Engine. It is part of Vertex AI Agent Builder.
• Purpose: To handle the infrastructure needed to scale agents in production, allowing developers to focus on creating applications.
• Core Services:
• Managed Runtime: Deploys and scales agents with end-to-end management capabilities. Supports customization of agent container images, security features (VPC-SC compliance, authentication, IAM), and access to models/tools like function calling. It supports agents built using ADK, LangChain, LangGraph, AG2, LlamaIndex, and custom frameworks.
• Context Management:
• Sessions (Preview): Stores individual interactions between users and agents, providing definitive sources for conversation context.
• Memory Bank (Preview): Stores and retrieves information from sessions to personalize agent interactions.
• Quality and Evaluation (Preview): Includes integrated Gen AI Evaluation service, Example Store for storing and dynamically retrieving few-shot examples, and optimization capabilities through Gemini model training runs.
• Observability: Provides tools to understand agent behavior via Google Cloud Trace (supporting OpenTelemetry), Cloud Monitoring, and Cloud Logging.
• Deployment Workflow: The typical workflow involves setting up the environment, developing the agent, deploying it to the managed runtime, using the agent via API requests, and managing the deployed agent.
• Security: Supports VPC Service Controls to strengthen data security and mitigate data exfiltration risks, blocking public internet access and confining data movement to authorized network boundaries.
• Streamlined Development: For an IDE-based experience, the agent-starter-pack offers pre-built agent templates (ReAct, RAG, multi-agent), an interactive playground, automated infrastructure (Terraform), CI/CD pipelines (Cloud Build), and built-in observability.

3. Agent2Agent Protocol (A2A)

• What it is: A new, open protocol launched by Google on April 9, 2025, with support and contributions from over 50 technology partners. It complements Anthropic’s Model Context Protocol (MCP).
• Goal / Vision: To enable AI agents to communicate with each other, securely exchange information, and coordinate actions across various enterprise platforms or applications. The aim is to allow agents to interoperate regardless of their vendor or framework, increasing autonomy and multiplying productivity gains. This universal interoperability is seen as essential for realizing the full potential of collaborative AI agents.
• Design Principles:
• Embrace agentic capabilities: Focuses on enabling agents to collaborate in their natural, unstructured modalities, not limiting an agent to a “tool”.
• Build on existing standards: Built on popular standards like HTTP, SSE, and JSON-RPC for easier integration with existing IT stacks.
• Secure by default: Designed to support enterprise-grade authentication and authorization, with parity to OpenAPI’s authentication schemes.
• Support for long-running tasks: Flexible to complete quick tasks or deep research taking hours/days, providing real-time feedback, notifications, and state updates.
• Modality agnostic: Supports various modalities, including audio and video streaming.
• How it Works: Facilitates communication between a “client” agent (formulating/communicating tasks) and a “remote” agent (acting on tasks).
• Capability Discovery: Agents advertise capabilities using an “Agent Card” in JSON format.
• Task Management: Communication is task-oriented, defined by the protocol, with output known as an “artifact”.
• Collaboration: Agents send messages to exchange context, replies, artifacts, or user instructions.
• User Experience Negotiation: Messages include “parts” with specified content types, allowing agents to negotiate needed formats and UI capabilities (e.g., iframes, video, web forms).
• Open Source: Google is committed to building the protocol in collaboration with partners and the community, releasing it as open source with clear pathways for contribution.
• Agentspace Connection: Google Agentspace supports the new open A2A Protocol, recognizing its criticality for multi-agent communication with a common language.

4. Google Agentspace

• What it is: A Google Cloud product launched in December, designed to put the latest Google foundation models, powerful agents, and actionable enterprise knowledge directly into employees’ hands.
• Core Purpose: To scale enterprise search and agent adoption. It helps employees and agents find information across the organization, synthesize and understand it with Gemini’s multimodal intelligence, and act on it using AI agents. It addresses critical enterprise needs for an AI-ready information ecosystem, easy agent creation, and enterprise-grade security/compliance.
• Key Features:
• Unified Agentic Search: Provides AI-powered multimodal search capabilities across diverse organizational data, including text, images, websites, audio, and video, regardless of where it’s stored (Google Workspace, Microsoft 365, Jira, Salesforce, ServiceNow, web content). It breaks down data silos by building an enterprise knowledge graph for each customer.
• Chrome Enterprise Integration (Preview): Allows employees to leverage Agentspace’s unified search capabilities directly from the search box in Chrome.
• Agent Gallery (Generally Available with allowlist): Offers a single view of available agents from Google, internal teams, and partners, making agents easy to discover and use within the enterprise. Partners can publish agents via Google Cloud Marketplace.
• Agent Designer (Preview with allowlist): A no-code interface for creating custom agents that connect to enterprise data sources and automate or enhance daily knowledge work tasks. Agents built in Vertex AI Agent Builder can be published to Agentspace.
• New Google-built Expert Agents:
• Deep Research Agent (Generally Available with allowlist): Explores complex topics, synthesizing information from internal and external sources into comprehensive reports from a single prompt.
• Idea Generation Agent (Preview with allowlist): Autonomously develops novel ideas in any domain and evaluates them to find optimal solutions.
• Agent2Agent (A2A) Protocol Support: Agentspace supports the new open A2A Protocol, which is critical for enabling multi-agent communication across different ecosystems.
• Enterprise-Grade Security: Built on Google’s secure infrastructure, it offers features like scanning for and blocking sensitive information (PHI, PII), role-based access controls, encryption with customer-managed keys, and data residency guarantees.
• AI Agent Marketplace: A dedicated section within Google Cloud Marketplace where customers can browse and purchase AI agents from partners (e.g., Accenture, Deloitte) to be made available in Agentspace.

Agent Landscape is continuously growing so its time to dive in rather than enjoying the shores.