Editor’s Note: This blog is adapted from a talk by Rushabh Vasa, Director at AMI and former Google Developer Expert for Cloud and AI at The Future of Finance Meetup at GeekyAnts. In this session, Rushabh breaks down the step-by-step process — from setting up the environment to training and deploying real-world AI agents.
Hi, I am Rushabh Vasa, Director at AMI, a cloud partner company. My team and I implement AI solutions for companies across industries worldwide. I previously served as a Google Developer Expert for Cloud and AI, and I have mentored startups through the Workforce Startup Accelerator. Over the last six years, I have seen AI evolve from theory to real-world impact.
Today, I will walk you through how to build AI Agents on Google Cloud using Vertex AI Agent Builder. This is the exact process I use with my teams to build real-world production-ready agents.
Introduction to AI Agents
Agents have taken center stage in the AI world. While traditional generative AI models generate content, agents go further — they take action. For example, if you ask ChatGPT to plan a trip, it generates an itinerary. An AI agent, on the other hand, generates the itinerary and books the flights and hotels automatically by connecting to external systems.
Agents combine reasoning, decision-making, and task execution, making them highly interactive and goal-oriented.
Setting Up Google Cloud Project
To get started, I created a Google Cloud Project inside the Google Cloud Console. After signing into my Google Cloud account, I created a new project, named it, and linked it to my billing account. With the project ready, I was able to move directly into the code lab interface provided by Google Cloud, which gives step-by-step instructions on building and deploying AI agents.
Starting with Google AI Agent Builder
Inside the Vertex AI Agent Builder, I began by selecting Create App. Google provides several agent templates, such as:
- Custom Search Agent: For building search experiences.
- Media Search Agent: For applying AI to media content.
- E-commerce Agent: For handling orders, recommendations, and customer conversations.
For this walkthrough, I selected Conversational AI Builder to create a chatbot-style agent.
Defining the Agent
Google allows you to build agents in two ways: conversational and chat-based. I chose the Conversational Agent option. After naming the project and setting time zone preferences, I activated Playbook Mode, which lets the agent follow instructions rather than a rigid flowchart.
For example, I set up an agent that assists users with travel-related queries. I added instructions like welcoming users, guiding them through destination options, and handling user queries fluidly. Google’s generative AI backend interprets these goals and drives conversations naturally, without requiring extensive programming.
Testing and Training the Agent
After the initial setup, I tested the agent’s responses by simply saying “Hi.” The agent greeted back, validating that it followed the instructions.
Next, I trained the agent with domain-specific data by connecting a Datastore. I uploaded documents in various formats — PDFs, text files, HTML, and structured JSON — to enrich its knowledge base. For example, while building a travel assistant, I uploaded documents about destinations, policies, and guidelines. The agent instantly started answering location-specific questions using this newly ingested data.
To refine accuracy, I configured data access behavior inside Agent Builder so the agent would prioritize this datastore when responding to relevant queries.
Real-World Use Cases
The flexibility of Vertex AI Agent Builder enables many practical use cases:
- Customer Support Agents that integrate with telephony and escalate to live agents.
- Internal Policy Agents that train on company handbooks to assist HR and finance teams.
- E-Commerce Agents that personalize recommendations and track orders.
- Support Bots that auto-train on website content to deliver instant responses.
With the ability to connect diverse data sources, agents can handle complex queries, deliver highly personalized experiences, and improve operational efficiency.
Publishing and Deployment
Once trained, I deployed the agent directly from the Agent Builder. Google provides simple integration code to embed the agent into websites or applications. The entire process — from setup to full deployment — can take less than an hour if your training data is ready.
The no-code interface eliminates the need for complex backend infrastructure or advanced programming skills, making it highly accessible for product and engineering teams alike.
Advanced Options
For teams that require more customization, Google Cloud supports advanced development workflows. The entire agent configuration can be exported, edited inside IDEs like VS Code, modified via Cloud Shell, or fine-tuned directly inside the cloud console.
Conclusion
With tools like Vertex AI Agent Builder, building real-world AI agents no longer requires deep machine learning expertise. This hands-on process allows anyone—from startups to enterprises—to create, test, and deploy intelligent agents quickly. As AI continues to evolve, agents like these are moving from early experimentation to real-world business applications faster than ever.
