The opinions expressed in this article are solely those of the author(Andrey Shakirov) and do not reflect the views of their employer.
This article explores the 2026 landscape of Hybrid Agentic Workflows on Google Cloud. We’ve moved beyond siloed choices to an integrated ecosystem powered by the Agent-to-Agent (A2A) protocol and the Model Context Protocol (MCP). Discover how to orchestrate a digital workforce where specialized agents collaborate across platforms.
TLDR: The Agentic Ecosystem
- Vertex AI Agent Builder - The Command Center. Unified governance and orchestration where you manage your digital employees.
- Gemini Enterprise - The Concierge. The employee-facing interface for grounded search, deep research, and agentic summaries. (Formerly Agentspace).
- Agent Development Kit (ADK) - The Engine. Code-first toolkit for custom logic, now supporting Python, Go, Java, and TypeScript.
- Google Antigravity - The Agentic IDE. The premier development environment where you build, deploy, and manage your agentic ecosystem using Gemini models.
This article outlines the key topics for building AI Agents on Google Cloud.
Vertex AI Agent Builder - low code, no code option
Google Antigravity - Agent Development IDE
Agent Development Kit - developer focused option
{#gemini-enterprise}
{#overview}
Gemini Enterprise serves as the primary touchpoint for employees to interact with the agentic ecosystem. In 2026, the paradigm has shifted from "Search Tool" to "Agent Management Hub," where employees supervise "digital assembly lines" of cooperating agents.
Here are the core pillars of Gemini Enterprise:
Deep Research & Idea Generation - Mature, out-of-the-box agents that can autonomously browse the web, analyze internal docs, and conduct tournament-style brainstorming.
- Deep Research: Gather, analyze, and synthesize cross-domain information with full citations.
- Idea Generation: Collaborative innovation framework powered by Gemini’s advanced reasoning.
Granular Grounding & Citations - Answers are now even more precise, with direct links to enterprise sections, ensuring high trust and verifiability.
Actions as Agent Manager - Employees don't just "query"; they trigger workflows. An employee's request in Gemini Enterprise can hand off tasks to custom ADK engines or Agent Builder playbooks via A2A.
Data Connectivity - Real-time reasoning over your operational data through native AlloyDB and BigQuery grounding integration.
App: A Gemini Enterprise app is an entity that delivers search results to your end users and if Gemini Enterprise Plus is enabled, an app also offers assistant and agent capabilities. The term app can be used interchangeably with the term engine in the context of APIs.
Data store: A data store is an entity that contains the data ingested from a first-party data source such as Cloud Storage or third-party applications such as Jira or Salesforce. Data stores that contain data from third-party applications are also called data connectors. Gemini Enterprise now integrates directly with AlloyDB for natural language querying over operational data.
Gemini Enterprise web application:
{#agent-gallery}
The Agents gallery is a portal in Google Gemini Enterprise that lets you access, create, and manage your agents. It showcases the following categories of agents:
- Premade by Google: Includes the agents that are available for you out-of-the-box and serve a specific function, such as Deep Research.
- Deep Research is a Premade by Google agent for users who need to gather, analyze, and understand internal and external information.
- Idea Generation is an agent that's premade by Google. Its goal is to help with innovation and problem-solving for enterprise users by combining advanced AI with a unique tournament-style competition framework.
- From your company: Includes the agents that your organization has created or has added.
- Your agents: Includes the no-code agents that you can create using Agent Designer.
{#build-new-agent-flow}
The Agent Designer in Gemini Enterprise lets you create your own agents, customized to your workflow and connected to your data—with no technical expertise required.
Once you click “Save” the new agent is deployed and production ready(all managed by Google Cloud) and ready to engage with your team in the Gemini Enterprise chat.
{#vertex-ai-agent-builder}
{#overview-1}
Vertex AI Agent Builder is the orchestration layer of the Google Cloud agentic ecosystem. It provides the governance, tools, and identity management needed to scale a fleet of digital employees across your organization.
{#cloud-api-registry}
Administrators can now curate and govern a central repository of tools. Whether it's Salesforce, SAP, or a custom BigQuery endpoint, tools are registered once and can be securely used by any authorized agent across Gemini Enterprise or custom ADK implementations using the Model Context Protocol (MCP).
{#agent-engine}
As of early 2026, several critical capabilities have reached General Availability:
- Code Execution: Agents can now securely execute code in sandbox environments to solve complex mathematical problems or process data on the fly.
- Memory Bank & Sessions: Persistent memory allows agents to remember user preferences and long-term context across multiple sessions.
- Agent Identity (IAM): Agents now have dedicated IAM-secured identities. They act as "digital employees" with specific permissions, ensuring they only access the data they are authorized to see.
Options to create a new Agent:
Left side menu with the options:
|
New Agent screen:
|
{#tools}
Using tools, you can connect playbooks to external systems. These systems can augment the knowledge of playbooks and empower them to execute complex tasks efficiently.
If you API requires authentication, you have following options to securely connect to it.
{#data-store-tools}
Generative agents can leverage data store tools to answer end-user questions using information from your data stores. Your list might have more options than what is displayed below.
{#connectors}
Connector tools can be used by an agent to perform actions using your Connections configured in Integration Connectors. Each connector tool is configured with a single Connection and one or more actions. If needed, multiple tools can be created for a single Connection to group different actions together for your agent to use.
{#function-tools}
Utilize function tools when client-side code offers functionality not supported by OpenAPI tools. These tools execute on the client side, not within the agent itself.
{#out-of-the-box-integrations}
This platform offers comprehensive integration capabilities, empowering agents for diverse customer interactions. "One-click Telephony" integrations like Avaya and Twilio streamline voice communication, enabling direct calling, routing, recording, and integrated call logs for efficient phone support.
For text-based interactions, the platform supports popular messaging channels such as Conversational Messenger, Facebook Messenger, LINE, Google Chat, Slack, Discord, Telegram, Viber, and even enterprise solutions like Facebook Workplace and Cisco Webex (formerly Spark).
Additionally, integrations with specialized messaging platforms like MMD Smart and services like Twilio (for general text messaging) and Azure Bot Service/Skype allow agents to engage with customers on their preferred channels, providing a seamless and centralized experience for handling inquiries across various digital communication touchpoints.
{#prebuilt-generative-agents}
Prebuilt Generative Agents offer a way to speed up the creation of new agents. They provide examples of solutions for typical tasks and situations, and can demonstrate recommended approaches.
{#integration-into-gemini-enterprise}
After creating your agent in Vertex AI Agent Builder, integrate it within the Gemini Enterprise application by navigating to Configurations / Assistant.
{#google-antigravity}
Google Antigravity is an agent-first Integrated Development Environment (IDE) designed for modern software development where AI agents are first-class citizens. Powered by Gemini models, it allows you to delegate entire tasks, not just lines of code.
{#agentic-paradigm}
Antigravity shifts development from manual coding to task orchestration. You provide a goal, and the agent:
- Plans: Breaks down the task into verifiable steps.
- Executes: Writes code, runs terminal commands, and manages dependencies.
- Verifies: Uses an integrated browser and test runner to ensure correctness.
{#key-features}
- 1M Context Window: Leverages Gemini 3.1 Pro to understand your entire project at once.
- Autonomous Actions: Agents can search the web, read documentation, and interact with your filesystem.
- Verifiable Artifacts: Generates task lists and implementation plans for human review, ensuring transparency.
- Multi-Agent Orchestration: Coordinate multiple agents working on different parts of your system simultaneously.
{#agent-builder-codelab}
Checkout this codelab that shows how to build GenAI Agent in Slack for Q&A over docs and actions with API calls: https://codelabs.developers.google.com/genai-for-dev-slack-agent
This codelab is based on the open source repository:
https://github.com/GoogleCloudPlatform/genai-for-developers
The repository contains reference implementations for different use cases like code reviews and dev tasks automation, etc.
{#agent-development-kit---developer-focused-option}
{#overview-2}
The Agent Development Kit (ADK) is the code-first framework for building custom agent logic. In 2026, ADK has expanded beyond Python to natively support Go, Java, and TypeScript, making it accessible to all enterprise backend teams.
{#gemini-live}
ADK now includes first-class support for Gemini Live integration. Agents built with ADK can handle bidirectional, low-latency audio and video streams, enabling "human-in-the-loop" visual troubleshooting and voice-guided workflows.
{#ecosystem-expansion} ADK's tool library has expanded to include deep, native integrations for:
- Productivity: Asana, Linear, Notion.
- Collaboration: Atlassian/Jira, Slack.
- Data: MongoDB, Redis, Pinecone.
{#mcp-default} ADK uses Model Context Protocol (MCP) by default, allowing tools to be shared between ADK agents and Vertex AI Agent Builder. It remains model-agnostic, supporting Gemini, Anthropic, and local VLLM endpoints.
{#adk-agents}
ADK offers various agent types designed to facilitate the creation of advanced applications. These categories provide a structured approach to development, enabling builders to leverage specific functionalities tailored to different AI use cases. Understanding these distinct categories is crucial for effectively utilizing the ADK to construct sophisticated and targeted solutions.
{#using-different-llm-models-with-adk}
The Agent Development Kit (ADK) is designed for flexibility, allowing you to integrate various Large Language Models (LLMs) into your agents. It is the perfect companion for Google Antigravity when building custom, high-performance agentic systems.
Here are some of the options for LLM models that ADK can use:
- Using Google Gemini Models (Gemini 3.1 Pro, Gemini 3 Flash)
- Using Anthropic models
- Using Cloud & Proprietary Models via LiteLLM
- Using Open & Local Models via LiteLLM
- Self-Hosted Endpoint (e.g., vLLM)
- Using Hosted & Tuned Models on Vertex AI
- Third-Party Models on Vertex AI (e.g., Anthropic Claude)
- etc
{#adk-tools}
Within the Agent Development Kit (ADK), a Tool equips AI agents with specific functionalities, allowing them to act and engage with the external world, going beyond standard text generation and logical reasoning. The effective utilization of tools is a key differentiator between advanced agents and fundamental language models.
From a technical perspective, a tool is usually a self-contained piece of code, such as a Python or Java function, a class method, or even another specialized agent. These tools are built to perform specific, predetermined tasks, frequently involving interactions with external systems or data sources.
Tools perform specific actions, such as:
- Querying databases
- Making API requests (e.g., fetching weather data, booking systems)
- Searching the web
- Executing code snippets
- Retrieving information from documents (RAG)
- Interacting with other software or services
{#adk-mcp-support}
ADK helps you both use and consume MCP tools in your agents, whether you're trying to build a tool to call an MCP service, or exposing an MCP server for other developers or agents to interact with your tools.
Here are couple of examples:
ADK Agent and FastMCP server deployed on Cloud Run with authentication enabled integration: https://github.com/gitrey/adk-fastmcp
MCP Toolbox for Databases is an open source MCP server that helps you build Gen AI tools so that your agents can access data in your database. Google’s Agent Development Kit (ADK) has built in support for The MCP Toolbox for Databases.
{#deployment}
Depending on your requirements for production readiness and the need for custom flexibility, your ADK agent offers deployment options across various environments.
{#agent-garden}
Agent Garden is a collection of ready-to-use samples and tools directly accessible within ADK. Leverage pre-built agent patterns and components to accelerate your development process and learn from working examples. GitHub repo: https://github.com/google/adk-samples
Leverage your team's existing expertise by building agents with popular open source frameworks like Agent Development Kit, LangGraph, or many more and then seamlessly deploying them on Vertex AI. Use Ironwood TPUs (v7) for massive scale and high-performance inference.
Connect custom agents to Google's enterprise-grade infrastructure without rewriting your code or changing your development workflow. Take advantage of Vertex AI's scaling, monitoring, and security capabilities while maintaining the flexibility of your chosen framework. Start with our step-by-step tutorials that guide you through the complete process from local development to production deployment.
{#conceptual-overview}
This 25-minute video provides a comprehensive overview of the ADK.
Agent Development Kit (ADK) - Conceptual Overview
{#testing-with-postman}
You can use your favorite tool postman to test your MCP server tools/endpoints:
{#real-world-use-cases}
Explore 601 real-world use cases from leading organizations leveraging AI Agents to improve their operations.
https://cloud.google.com/transform/101-real-world-generative-ai-use-cases-from-industry-leaders
Customer Agents: Focusing on customer interaction and service (e.g., United Wholesale Mortgage, Motorola, Personal AI).
Employee Agents: Assisting with internal processes and employee productivity (e.g., Toyota, Uber, Flashpoint).
Code Agents: Utilizing AI for software development and related tasks (e.g., Capgemini, CME Group, ROSHN Group).
Data Agents: Managing and interpreting data (e.g., Colombia's Ministry of Information and Communications Technologies, BT Group).
Creative Agents: Supporting creative endeavors (e.g., The Estee Lauder Companies).
Security Agents: Detecting fraud and combat money laundering(e.g., Airwallex, Bradesco)
{#conclusion}
Developing AI Agents on Google Cloud is no longer about choosing a single tool. It is about building an agentic workforce where:
- Gemini Enterprise provides the friendly Concierge for your employees.
- Vertex AI Agent Builder acts as the Command Center for governance and A2A orchestration.
- Agent Development Kit (ADK) provides the specialized Engine for custom, high-scale logic.
- Google Antigravity serves as the Development IDE to wire it all together.
By leveraging A2A and MCP, your agents move beyond siloes, collaborating securely to transform your business processes.
This is series of articles on building and deploying applications on Google Cloud that unpack discussed options in more details.
-
Design, Prototype, Build, and Deploy on Google Cloud: A Comprehensive Guide
-
So You've Built Your Idea and Want to Share It with the World?
* The opinions expressed in this article are solely those of the author(Andrey Shakirov) and do not reflect the views of their employer.