Vertex AI & Gemini Enterprise Agent Platform

By the end of this lesson you will be able to explain Google Cloud's AI platform and its 2026 name change, call a Gemini model from Python, and know which part of the platform to use for chatbots, document search, agents and classic machine learning. You will also learn the safety and data habits that matter for Indian users.

A name change you must know

For years, Google Cloud's machine learning platform was called Vertex AI. On 22 April 2026 Google announced that Vertex AI is now part of the Gemini Enterprise Agent Platform. Many features were renamed at the same time. You will meet both sets of names in blogs, courses and job descriptions:

The underlying API service is still enabled as aiplatform.googleapis.com, and the gcloud commands are still under gcloud ai. In this lesson, "the platform" means Gemini Enterprise Agent Platform, formerly Vertex AI.

What the platform offers

Think of it as a large hospital. Different departments handle different needs, but they share one reception (your project), one records room (your data) and one security desk (IAM).

• Gemini models: Google's generative AI models for text, code, images, audio and video, called through an API.
• Model Garden: a catalogue of Google, partner and open models you can try, tune or deploy.
• Agent Studio: a console playground to test prompts and settings before writing code.
• Agent Search and RAG Engine: answer questions from your own documents. RAG (retrieval-augmented generation) means the app first finds relevant passages, then asks the model to answer using only them.
• Agent Runtime: a managed place to deploy and run AI agents built with frameworks such as Google's Agent Development Kit.
• Classic ML: AutoML, custom training, a Model Registry and endpoints for your own models, such as predicting which students may drop out.

Step by step: call Gemini from Python

Step 1. Enable the API in your project and sign in for local code:

gcloud services enable aiplatform.googleapis.com
gcloud auth application-default login
pip install --upgrade google-genai

Step 2. Write ask.py. The Google Gen AI SDK (google-genai) is the current Python library for Gemini. enterprise=True tells it to use the Agent Platform in your Google Cloud project, with your IAM permissions and billing:

from google import genai
from google.genai import types

client = genai.Client(enterprise=True, project="jaipur-coaching-dev", location="global")

response = client.models.generate_content(
    model="gemini-3.5-flash",
    contents="Explain photosynthesis to a Class 7 student in three short sentences.",
    config=types.GenerateContentConfig(
        system_instruction="You are a patient science teacher in India. Use simple English.",
        temperature=0.2,
    ),
)
print(response.text)

Step 3. Run it with python ask.py. You will see something like this:

Plants make their own food using sunlight, water and carbon dioxide.
This happens mainly in the green leaves, which contain chlorophyll.
As a result, plants give out oxygen, which we breathe.

Model answers vary from run to run, so your wording will differ. A low temperature makes answers more focused and repeatable; a higher one makes them more creative. The system instruction sets the model's role and style for every request.

The model ID gemini-3.5-flash was listed as generally available when this course was checked. Model names change often, so check the current model list in the platform documentation before you build. Pricing is usually per amount of input and output (tokens); see the official pricing page rather than guessing.

Building something useful

Three common patterns, from simple to advanced:

1. Prompt plus your data: put a short document (such as a fee policy) in the prompt and ask questions about it. Fine for small documents.
2. RAG over many documents: index your college handbook, circulars and FAQs with RAG Engine or Agent Search. The app retrieves relevant parts and the model answers with citations. This reduces made-up answers, called hallucinations.
3. Agents: an agent can call tools, such as "look up a student's fee status" through your API, decide the next step, and finish a task. Deploy it on Agent Runtime or Cloud Run.

For classic ML, the workflow is: prepare data (often in BigQuery), train with AutoML or custom code, register the model, deploy it to an endpoint, and send prediction requests. You can see deployed endpoints with gcloud:

gcloud ai endpoints list --region=asia-south1

Using endpoint [https://asia-south1-aiplatform.googleapis.com/]
Listed 0 items.

On a new project there are none yet, which is what this output shows.

Using AI responsibly

AI features need the same care as any other system, and a few extra rules:

• Do not put secrets or unnecessary personal data in prompts. Remove names, phone numbers and Aadhaar numbers where they are not needed.
• Check data location. The global location used above lets Google route requests to any available region. If data must be processed in India, check which models and locations support that, and follow Data Residency & Compliance in India.
• Keep a human in the loop for decisions that affect people, such as admissions, marks or loans.
• Evaluate before launch. Build a small test set of real questions with good answers, and measure the model's answers against it after every change.
• Use safety settings and grounding. Filter harmful content and ground answers in trusted documents.
• Use service accounts, not personal logins, when the code runs on Cloud Run or elsewhere in production.

Common mistakes

• Following 2025 tutorials blindly. Names, SDKs and model IDs have changed; check the current docs.
• Hard-coding an old model ID. Keep the model name in configuration so you can upgrade easily.
• Sending personal data to a global endpoint when rules require Indian processing.
• Trusting answers without checks. Ground important answers in your documents and test them.

Key points

• Vertex AI became part of the Gemini Enterprise Agent Platform in April 2026, and many features were renamed.
• The Google Gen AI SDK (google-genai) calls Gemini models; the client uses your project, IAM and billing.
• Model Garden, Agent Studio, RAG Engine, Agent Search and Agent Runtime cover models, testing, document answers and agents.
• Classic ML still uses training, a model registry and endpoints.
• Protect personal data, check processing locations, evaluate outputs and keep humans in important decisions.

from google import genai
from google.genai import types

POLICY = (
    "Fees are due by the 10th of each month. "
    "A late fee of 100 rupees applies after the 15th. "
    "Siblings get 10 percent off."
)

client = genai.Client(enterprise=True, project="jaipur-coaching-dev", location="global")
response = client.models.generate_content(
    model="gemini-3.5-flash",
    contents=f"Policy:\n{POLICY}\n\nQuestion: My fee is late by 7 days. Do I pay a late fee?",
    config=types.GenerateContentConfig(
        system_instruction="Answer only from the policy. If the policy does not say, reply: I don't know.",
        temperature=0,
    ),
)
print(response.text)