This is the largest domain, about 35% of the exam. The main skill is matching a business need to the right Google product and knowing the one-line job of each. Names changed in 2026: Vertex AI is now Gemini Enterprise Agent Platform (Agent Platform), Vertex AI Studio is Agent Studio, Vertex AI Search is Agent Search, and Agentspace became Gemini Enterprise. The current guide uses the new names. Old material and old questions may use the old ones.
Google Cloud’s strengths in gen AI
| Strength | What it means |
|---|---|
| AI-first approach | Google has researched and shipped AI for over a decade (the Transformer paper, TPUs, DeepMind, Search). It runs AI in its own products at huge scale and brings the lessons to customers |
| Enterprise-ready platform | Responsible, secure, private, reliable and scalable. Customer data is not used to train Google’s models without permission, with encryption, IAM, data residency and compliance controls |
| Comprehensive ecosystem | Gen AI is built into products people already use: Workspace, Search, BigQuery, Chrome, Android, Looker, Security tools. Integration shortens time to value |
| Open approach | Support for open models (Gemma, third-party open models), open standards and frameworks, a choice of models in Model Garden, multi-cloud and open source tooling. Lowers lock-in |
| AI-optimized infrastructure | Custom chips and a purpose-built stack, below |
| Control over data | Security, privacy and governance features, plus a choice of first-party and open models, prebuilt and customizable solutions, and agents |
| Democratized development | Low-code and no-code tools, pre-trained models and APIs let non-experts build |
AI-optimized infrastructure
- AI Hypercomputer is Google’s integrated supercomputing system for AI: hardware, software and consumption options designed together so training and serving are fast and cost-efficient.
- TPUs (Tensor Processing Units) are Google’s custom-designed chips for machine learning. Strong price-performance and efficiency for training and serving large models.
- GPUs from partners such as NVIDIA are also available, for teams that want that ecosystem.
- Underneath: Google’s global data centres, private fibre network, and efficient cooling and power. Benefit to a business: speed, scale, lower cost per result and a sustainability story.
Trap: TPUs and GPUs are not alternatives you must choose in advance. Google Cloud offers both. TPUs are the Google-designed option.
Control over data
Google Cloud’s platform gives customers: encryption by default and customer-managed keys, IAM and least privilege, VPC Service Controls to limit data exfiltration, data residency choices, audit logs, and a commitment that your prompts and data stay yours and are not used to train foundation models for others without your permission.
Democratizing AI development
- No-code and low-code: Gemini Enterprise agent designer, Conversational Agents, Agent Studio’s visual tools, Google AI Studio.
- Pre-trained models and APIs: Vision, Speech, Translation, Natural Language, Document AI. No ML expertise needed.
- AutoML: train a custom model from your own labeled data without writing model code.
Prebuilt gen AI for AI-powered work
| Product | For whom | What it does |
|---|---|---|
| Gemini app | Individuals (consumer, and Workspace users) | Google’s AI assistant for chat, writing, image creation, research and coding help. Paid tiers (formerly “Gemini Advanced”, now under Google AI plans) give access to the most capable models and larger limits |
| Gems | Gemini app users | Custom versions of Gemini you create with saved instructions and files for a repeatable task (a writing coach, a meeting-prep helper) |
| Gemini Enterprise | Employees across a company | An enterprise agentic platform with a front door for work: multimodal search across company data, connectors to systems such as Workspace, Microsoft 365, Salesforce and others, prebuilt agents (for example deep research), a no-code agent builder for custom agents, and notebook-style synthesis of documents. Respects existing permissions |
| Gemini for Google Workspace | Workspace users | Gemini inside Gmail, Docs, Sheets, Slides, Meet, Drive and Chat: draft, summarise, rewrite, take meeting notes, analyze sheets, create slides |
Choosing between them:
- One person wants a personal assistant: Gemini app.
- Staff need AI inside the tools they already use for documents and email: Gemini for Workspace.
- Staff need to search and act across all company systems and have custom agents: Gemini Enterprise.
- A developer builds an AI feature into a product: Agent Platform (below), not these.
Business value: productivity, faster information discovery, less context switching, and no need for a development team.
Gen AI that improves customer experience
External search
- Agent Search (on Gemini Enterprise Agent Platform; formerly Vertex AI Search) builds Google-quality search and answer experiences over your own content, for websites, apps and customer portals. It can return ranked results plus generated answers with citations.
- Grounding with Google Search lets a Gemini model use live Google Search results to give current, sourced answers (covered in the grounding domain).
Customer Engagement Suite
A set of products for contact centres and digital channels, powered by Gemini.
| Product | Job |
|---|---|
| Conversational Agents | Build chat and voice agents for customers, in text or phone, with no-code flows plus generative answers grounded in your data. Replaces rigid menu bots |
| Agent Assist | Helps human agents live: suggests answers, summarises the call, finds knowledge articles, drafts replies |
| Conversational Insights | Analyzes conversations at scale: topics, sentiment, why customers call, quality and compliance |
| Google Cloud Contact Center as a Service (CCaaS) | A full cloud contact centre platform with the AI built in: routing, channels, agents and analytics |
Benefits: shorter waits, 24/7 service, consistent answers, lower handling time, better insight from every call. Handing complex cases to people with context is a design goal.
Empowering developers: Agent Platform
Gemini Enterprise Agent Platform is Google Cloud’s unified platform to access models, build agents, evaluate, deploy and govern at enterprise scale.
| Component | What it does |
|---|---|
| Model Garden | A catalogue of hundreds of models: Google’s Gemini, Imagen, Veo and Gemma, third-party models (for example Anthropic Claude) and open models. Try, compare, tune and deploy from one place |
| Agent Studio | A console to prototype and test prompts and models, tune models and build agents, with enterprise security and governance (formerly Vertex AI Studio) |
| Agent Search | Enterprise search and ready-made RAG over your data |
| AutoML | Train custom ML models on your own labeled data with little or no code (images, text, tabular, video) |
| Custom training and tuning | Fine-tune foundation models or train your own with your data |
| Agent building | Build custom agents with an open-source Agent Development Kit (ADK), a managed runtime to deploy and scale them, and no-code builders |
| Evaluation and monitoring | Compare model outputs, score quality, watch for drift |
| Feature Store, Model Registry, Pipelines | MLOps pieces for managing data features, versions and workflows |
Why choose Agent Platform: model choice, customisation on your data, enterprise security and governance, integration with BigQuery and other Google Cloud data, and one place from experiment to production.
Google Cloud’s RAG offerings
Retrieval-augmented generation lets a model fetch your facts before answering. Google offers it at three levels of effort.
| Offering | Effort | Use when |
|---|---|---|
| Pre-built RAG with Agent Search | Lowest. Point it at your content (websites, Cloud Storage, BigQuery, connectors) and it indexes, retrieves and answers | You want a working Q&A or search experience quickly without building a pipeline |
| RAG APIs (for example the RAG Engine and the grounding and ranking APIs) | Medium. Developer controls chunking, retrieval, ranking and prompts and plugs into their own app | You need custom control but not a whole pipeline from scratch |
| Build your own RAG with a vector store and embeddings | Highest | Special requirements that managed options can’t meet |
Custom agents on Agent Platform
An agent uses a model to reason, plan and use tools to reach a goal. Use Agent Platform to build agents when off-the-shelf products do not fit: your own workflows, systems and brand. Gemini Enterprise’s agent builder serves business users; ADK and Agent Platform serve developers who need code-level control.
Tooling for agents
Agents act by calling tools. The guide names four kinds of tool.
| Tool type | How it works | Example |
|---|---|---|
| Extensions | Standard bridge from the agent straight to an external API. The agent decides when to call and the extension runs it | Look up flight status through an airline API |
| Functions | The model proposes a call with the right arguments. Your application code runs it on the client side and returns the result | Model asks to call get_order_status(order_id), your code executes it |
| Data stores | Give the agent access to your documents or databases so it can retrieve facts (RAG) | Product manuals in a search index |
| Plugins | Add-ons that extend an agent with a packaged capability | Calendar or payment capability added to an agent |
Extension vs function: with an extension the agent side executes the API call; with a function the developer’s app executes it (more control, useful for security checks, rate limits, or when the API is not reachable by the agent).
Modern agent tooling also uses open protocols such as MCP (Model Context Protocol) for tools and A2A (Agent2Agent) for agent-to-agent communication. They are not in the guide’s list but give context for “open approach”.
Google Cloud services and AI APIs as agent tools
| Service | Role as a tool |
|---|---|
| Cloud Storage | Holds files and documents the agent reads or writes |
| Databases (BigQuery, Cloud SQL, Spanner, Firestore, AlloyDB) | Structured data for lookups and updates |
| Cloud Functions / Cloud Run | Host the code or microservices that agent functions call |
| Agent Platform | Models, search, RAG and agent runtime |
| Speech-to-Text API | Turn audio into text |
| Text-to-Speech API | Turn text into natural speech |
| Translation API | Translate text between languages |
| Document Translation API | Translate whole documents and keep formatting |
| Document AI API | Extract structure and fields from forms, invoices, contracts and IDs |
| Cloud Vision API | Labels, objects, text (OCR), faces and unsafe content in images |
| Cloud Video Intelligence API | Detect objects, scenes and text in video |
| Natural Language API | Sentiment, entities and categories in text |
| Google Cloud API Library | The catalogue of all Google Cloud APIs you can enable and call |
These are pre-built AI APIs: no model building, call them like any web service. Pick them when you need one narrow, proven capability.
Agent Studio vs Google AI Studio
| Agent Studio (Agent Platform) | Google AI Studio | |
|---|---|---|
| Audience | Enterprise teams on Google Cloud | Individual developers and small teams |
| Access | Google Cloud project, IAM | Google account and an API key, quick start |
| Strengths | Governance, security, data controls, tuning, evaluation, deployment, Model Garden | Fastest way to try the latest Gemini models and prototype |
| Pick when | Moving toward production with enterprise needs | Experimenting and prototyping quickly |
A common path is prototype in AI Studio, then move to Agent Studio for production.
Quick product matching
| Need | Pick |
|---|---|
| Personal AI assistant, custom helpers | Gemini app and Gems |
| AI in Gmail, Docs, Meet | Gemini for Workspace |
| Search and agents across all company data for employees | Gemini Enterprise |
| Customer-facing chat or voice bot | Conversational Agents |
| Help live human agents on calls | Agent Assist |
| Learn from call transcripts | Conversational Insights |
| Whole contact centre in the cloud | Google Cloud CCaaS |
| Search or Q&A on your website or content | Agent Search |
| Try and compare many models, tune, deploy | Agent Platform and Model Garden |
| Train custom model on labeled data, little code | AutoML |
| Extract fields from invoices | Document AI |
| Quick prototype with an API key | Google AI Studio |
| Run models on Google-designed chips | TPUs on AI Hypercomputer |
Common traps
- Gemini Enterprise is for employees; Customer Engagement Suite is for customers and contact centres.
- Agent Search is a ready-made RAG/search product; RAG APIs are building blocks.
- Document AI extracts data from documents; Vision API reads images generally.
- AI Studio is a quick-start tool; production governance belongs in Agent Studio on Google Cloud.
- A prebuilt API is chosen for a narrow task. A foundation model is chosen for open-ended generation.
Exam checklist
- List Google Cloud’s strengths: AI-first, enterprise-ready, ecosystem, open approach, optimized infrastructure, data control, democratization.
- Explain AI Hypercomputer, TPUs and GPUs in a sentence.
- Tell the Gemini app, Gems, Gemini Enterprise and Gemini for Workspace apart.
- Name the four Customer Engagement Suite parts and their jobs.
- Describe Agent Platform: Model Garden, Agent Studio, Agent Search, AutoML, agent building.
- Rank RAG options by effort: prebuilt with Agent Search, RAG APIs, build your own.
- Explain extensions, functions, data stores and plugins, and extension vs function.
- Match each pre-built AI API to its input and output.
- Choose between Agent Studio and Google AI Studio.