Generative AI Leader · §2 · ~35 %
Section 2: Google Cloud's gen AI offerings
· approx. 5 min read
Independent study aid – not by Google, not affiliated with Google.
The largest exam section asks which Google offering fits which need: the Gemini app, Gemini for Google Workspace, Gemini Enterprise and Gemini Notebook for everyday work, the Customer Engagement Suite for customer service, Gemini Enterprise Agent Platform with Model Garden and Agent Studio for developers, and tools and APIs for agents. Weight in the exam: approx. 35 %.
2.1 Google Cloud's strengths in gen AI
- AI-first company: For decades, Google has combined ML research with products. Publicly known milestones: its own TPUs (from 2016), the Transformer paper "Attention Is All You Need" (2017), BERT (2018), the Google AI Principles (2018), the Bard chatbot (2023), the Gemini model family (Dec. 2023) and the open Gemma models (2024). The Study Guide summarizes the benefit like this: gen AI tools are integrated across the entire Google ecosystem, and Google keeps customers at the forefront of AI development. Those who use Google Cloud benefit from these advancements without starting from scratch.
- Enterprise-ready: responsible, secure, private, reliable, scalable. Concretely: security infrastructure (encryption, access control via IAM, network security), compliance certifications, and an Agent Platform with enterprise-grade security, scalability and governance. The Study Guide speaks of an "enterprise-grade foundation you can build on" and an ecosystem that puts security and ethics first.
- Data stays under the customer's control: Gemini for Google Cloud does not use prompts and responses to train its models; Google Cloud's standard safeguards apply. With Gemini Notebook Enterprise, the data stays in the customer's own Google Cloud project. The Exam Guide lists these building blocks of data control: security, privacy, governance, open and leading first-party models, prebuilt and customizable solutions, agents.
Common exam trap
Absolute promises ("guaranteed", "100 % secure", "complete accuracy", "ready-made solution for every industry") are almost always wrong. Reason: the Study Guide itself names limitations of foundation models (hallucinations, bias, knowledge cutoff, edge cases) and recommends human-in-the-loop for critical decisions. A serious provider therefore cannot promise error-free results.
2.2 Prebuilt gen AI offerings for AI-powered work
- The Gemini app (gemini.google.com, mobile) is Google's gen AI chatbot. It helps with writing, summarizing, translating and generating images.
- Gemini is the model, the Gemini app is the application. The Study Guide puts it this way: Gemini is a model that powers many different solutions. Gen AI itself is a technology, not an application.
- Gemini Advanced gives businesses additional features and enterprise-grade protection. Typical are access to the most capable models (strong at coding, logical reasoning, nuanced instructions), larger context windows and early access to new features.
Common exam trap
Gemini app vs. Gemini Enterprise vs. Gemini for Workspace:
2.3 Gen AI offerings that improve the customer experience
- Agent Search on Gemini Enterprise Agent Platform (formerly: Vertex AI Search) provides search and recommendation solutions for your own business, i.e. Google-quality search over your own data.
- Google Search as an offering: Grounding with Google Search connects Gemini to current web information. This helps against the knowledge cutoff, i.e. the limit of the training knowledge, e.g. for questions about current events, prices or results.
- Tools that let businesses interact effectively with customers. They can run on top of Google Cloud Contact Center as a Service (CCaaS).
Common exam trap
Agent Assist vs. Conversational Agents:
2.4 How Google Cloud empowers developers to build with AI
- Google Cloud's unified ML platform for building, training and deploying ML applications. It provides access to gen AI models (e.g. Gemini), lets you tune and deploy them, and supplies infrastructure, tools and pre-trained models for the entire lifecycle. It is fully managed: you configure, Google operates. According to the docs, the platform is an evolution of Vertex AI; it was renamed on 2026-04-22.
- Characteristics: open and flexible, powerful infrastructure, pre-trained models, comprehensive tools, customization, easy integration.
- The public docs (2026) organize the platform into Build (ADK, Agent Studio, Agent Garden, Model Garden, RAG Engine, Vector Search), Scale (Agent Runtime, Sessions, Memory Bank), Govern (Agent Registry, Agent Identity, Agent Gateway, Policies) and Optimize (Evaluation, Observability, prompt optimization).
Common exam trap
Model Garden vs. AutoML vs. custom training:
- Ready-made, capable model, fast, moderate budget → Model Garden.
- Your own model on your own labeled data without ML expertise, e.g. sales forecasting from tables → AutoML.
- Full control over architecture, research, special requirements → Custom training.
2.5 Purpose and types of tooling for gen AI agents
- Agents use tools to interact with their environment, e.g. to retrieve and process data or trigger actions. Without tools, a model remains limited to its training knowledge. Tools are also the way agents collaborate with other agents.
- Four tool types:
- Extensions: connect to external services via APIs and standardize the interaction despite differing API designs.
- Functions: define specific actions or tasks. Reusable building blocks that the reasoning system selects as appropriate. With function calling, the model produces a structured call (name + parameters); it is executed by the application, which returns the result.
- Data stores: provide access to information (current, historical, knowledge base) for accurate and relevant answers. The foundation for RAG.
- Plugins: add new skills and integrations, e.g. connections to other platforms and specialized tools.
- Reasoning loop with tools: (1) reasoning, i.e. tool selection → (2) acting, i.e. executing the tool → (3) observation, i.e. receiving the result → (4) iteration, i.e. planning the next step. Example: an agent at a carpentry business is to schedule an on-site measurement appointment. It reads customer data from a data store, checks free slots via a calendar plugin, calls a "create appointment" function and confirms via an email extension.
Common exam trap
Agent Studio vs. Google AI Studio: The difference lies in target audience and maturity: rapid, free prototyping vs. production-ready applications at scale. Not image vs. text, not code vs. no-code, not custom vs. pre-trained models.
Read next: All exam sections · Glossary · Product map