Google Cloud's generative AI offerings, mapped

Google sells generative AI under a lot of names, and several of them start with "Gemini". There's a Gemini model, a Gemini app, Gemini inside Gmail and Docs, Gemini Enterprise, and Gemini Enterprise Agent Platform, all offered at once. On top of that the names keep changing: Vertex AI became Gemini Enterprise Agent Platform in April 2026, Google Agentspace became part of Gemini Enterprise, and NotebookLM Enterprise became Gemini Notebook Enterprise.

The Generative AI Leader exam gives this lineup the largest share of its weight, about 35%, and most of the questions read the same way. A business has a need, and four real Google products are offered as answers. Usually only one is aimed at the right people at the right level of effort, and the other three are real products applied to the wrong problem. This guide sorts the lineup into a map you can use to answer those questions.

Two questions that sort almost everything

Before you look at product names, ask two things about the scenario.

Who uses it? Employees doing their own work, customers of the business, or developers building something new. The exam guide is organised the same way: prebuilt offerings that "enable AI powered work", offerings that "improve the customer experience", and the ways Google Cloud "empowers developers to build with AI".
How much does the business want to build? Nothing (turn it on and use it), some configuration (connect data sources, describe an agent in plain language), or a custom application with code, its own data pipeline and its own deployment.

Once you have both answers, there is usually only one product left.

For employees: ready-made apps

The Gemini app

The Gemini app is the general-purpose assistant: chat, writing, analysis, image generation and Deep Research. Its feature most likely to come up on the exam is Gems, custom versions of Gemini that save a set of instructions and reference files, so a task someone repeats every week needs only a short request. Building a Gem takes no code. Google says the Gemini app is included at no extra cost for Google Workspace commercial and public sector customers, with admin controls from the Workspace console.

Pick it when one person or a team needs a capable assistant for general tasks, and the answer doesn't depend on searching across the company's other systems.

Gemini for Google Workspace

This is Gemini built into the apps people already use: drafting in Gmail and Docs, building tables in Sheets, taking notes in Meet. The point is that nothing changes about where people work, and nobody pastes confidential text into a separate tool.

The permission model matters, and the exam tests it. Gemini has the same access to Workspace data as the user does: it can only use the Drive files shared with that person and the calendar events they can see. If a file's owner has turned off download, copy and print, Gemini can't use that file even though it is shared. Turning Gemini on does not widen anyone's access.

Pick it when the organisation runs on Workspace and the goal is help inside everyday documents, mail and meetings.

Gemini Enterprise

Gemini Enterprise is Google's intranet search, AI assistant and agent platform for employees, and it is the product that absorbed Google Agentspace. The key difference from Workspace is reach. It comes with prebuilt connectors for systems beyond Google Workspace, such as Microsoft SharePoint and OneDrive, Jira, Confluence and ServiceNow, and its search is permissions-aware, so each employee only sees results and answers drawn from content they could already open.

It also carries the agent features: ready-made agents in the Agent Gallery, a no-code builder for business users (now called Workflow Builder, formerly Agent Designer), and a way to publish agents that developers built on Agent Platform with the Agent Development Kit. Gemini Notebook Enterprise, for research grounded in documents a user uploads, is sold alongside it.

Editions matter less on the exam than the idea, but for reference Google lists a Business edition for teams of up to 300 with no IT setup, Standard and Plus editions for organisations with stricter security and compliance needs, and a Frontline edition for frontline workers. Google also states that customer prompts and outputs in the Business, Standard and Plus editions aren't used to train Google's models.

Pick it when the need is "one place where staff can ask questions across all our systems", or when business teams want to build and share their own agents without a development project.

For customers: search and the contact center

Agent Search

Agent Search on Gemini Enterprise Agent Platform, formerly Vertex AI Search, builds Google-quality search over a company's own websites and its structured and unstructured data. A retailer whose site search only matches exact keywords can move to search that understands a query like "warm jacket for rainy hikes", sometimes by little more than adding a search widget.

Agent Search has a second job: it is Google's out-of-the-box RAG system. It takes care of chunking, embedding, indexing and retrieval, so a generative AI app can be grounded in company documents without anyone building a retrieval pipeline. The exam guide lists "pre-built RAG with Agent Search" next to the more configurable RAG APIs and Grounding with Google Search.

Don't confuse it with Gemini Enterprise. Both search company content, but Agent Search is a building block that developers put into an app or website, often one customers use. Gemini Enterprise is a finished app for employees.

The contact center products

The exam guide calls this the Customer Engagement Suite: Conversational Agents (virtual agents that talk to customers directly), Agent Assist (live suggestions, summaries and coaching for human agents during a conversation), Conversational Insights (analysis of past conversations for sentiment, topics and quality) and Google Cloud Contact Center as a Service. Google's documentation now groups this area under Gemini Enterprise for Customer Experience, where the analytics product appears as Customer Experience Insights.

The trap here is who is being helped. If the customer talks to the AI, it's a conversational agent. If a human agent gets help while talking to the customer, it's Agent Assist. If managers want patterns across thousands of calls after they've happened, it's Insights.

For developers: Gemini Enterprise Agent Platform

Gemini Enterprise Agent Platform is what Vertex AI is now called. Google describes it as an evolution of Vertex AI: one platform to build, deploy, govern and optimise models and agents. The parts the exam names most often:

Model Garden, a library to discover, test, customise and deploy models from Google, from partners and from the open model ecosystem, all in one place with one deployment pattern.
Agent Studio (formerly Vertex AI Studio), the low-code workspace for trying models, designing prompts and building agents without code.
Agent Development Kit (ADK), the open-source, code-first framework for multi-agent systems where developers want full control over reasoning and tools, and Agent Runtime (formerly Agent Engine), the managed environment that hosts them.
Agent Search and RAG Engine for grounding answers in company data.
Training and tuning tools such as AutoML for teams that need a custom model.

Agent Studio or Google AI Studio?

The exam guide asks this one directly. Google AI Studio and the Gemini API are the fastest way to start, with a Google Account and an API key. Agent Platform is the enterprise path, which needs a Google Cloud account with billing. Google's own comparison says Agent Platform supports compliance with certifications such as HIPAA and SOC 2, and offers data residency, customer-managed encryption keys, Access Transparency, IAM-based authentication instead of API keys, VPC security, enterprise support and availability SLAs. The Gemini API side lists none of those. So a prototype in AI Studio is fine, but anything that handles regulated data or needs an SLA belongs on Agent Platform.

Build or buy

Underneath all of this is the build-versus-buy decision the exam keeps returning to. Prebuilt offerings (the Gemini app, Gemini for Workspace, Gemini Enterprise) are quick to deploy, need no engineering team and cover common needs well. Building on Agent Platform costs more time and skill, but it's the only route when the AI must sit inside a proprietary workflow, call internal systems, or reach customers in a way no packaged product does.

If a scenario says "no development team" or "running within weeks", look for the prebuilt answer. If it says "full control", "our own models" or "integrated with our internal systems", look for Agent Platform.

A worked example

A logistics company has three requests on the CIO's desk.

Sales staff want help drafting proposals and summarising long email threads. The company runs on Gmail and Docs. Gemini for Google Workspace: the help appears where they already work, and it respects existing sharing.
Operations staff can't find answers that are spread across SharePoint, Jira and ServiceNow, and there's no budget for developers. Gemini Enterprise: connectors plus permissions-aware search, with no application to build.
The engineering team wants a customer-facing assistant that checks shipment status in the company's own tracking system and books redeliveries. Agent Platform: an agent built with the ADK, grounded with Agent Search, calling the tracking API as a tool and hosted on Agent Runtime.

Swap any two of these answers and the result is either far too much engineering or a tool that can't reach the data it needs.

Traps to watch for

A developer platform offered for a need that a prebuilt app already covers. If nobody is going to write code, Agent Platform is the wrong answer.
Gemini for Google Workspace offered when the data lives outside Workspace. Data in Jira or ServiceNow points to Gemini Enterprise.
Agent Search and Gemini Enterprise confused. Ask whether it is a component for a customer-facing app or a finished app for employees.
Old names. Questions written for today's exam use Agent Platform, Agent Studio and Agent Search, but study material from 2025 says Vertex AI, Vertex AI Studio and Vertex AI Search. They're the same products.