AIAI Agent Dev Labaiagentdev-lab.com
Sign in to save progressSign in
Menu

GAIL · LESSON 2

Section 2: Google Cloud's generative AI offerings約35% · 5 topics

Learn Google Cloud's strengths, ready-to-use generative AI, improving customer experience, developer services, and agent tools.

2.1Google Cloud's strengths

Google Cloud's strengths (AI-first approach, enterprise readiness, ecosystem, open approach, AI Hypercomputer, TPUs and GPUs, data management, democratizing AI with low-code/no-code)

Concept

Google Cloud's generative AI strengths are AI-first R&D, enterprise security and data protection, a partner ecosystem, an open approach that lets you choose other companies' and open models, AI Hypercomputer (AI compute combining TPUs and GPUs), data management such as BigQuery, and democratizing AI through low-code/no-code tools.

How it works / steps

  1. Clarify the requirements you value (security, choice, data integration)
  2. Compare not only Google's models but also partner and open models in Model Garden
  3. Use low-code tools so non-engineers can use AI too

Workplace example

A company that also wants other vendors' models compares Gemini and others on the same platform in Model Garden, while keeping its data consolidated in BigQuery.

Common confusions and exam tips

Mini quiz

Q1. Which best describes AI Hypercomputer?

  1. An API for adding digital watermarks to generated images
  2. An employee app for searching internal documents
  3. An AI computing platform integrating TPUs, GPUs, storage and more
  4. A dedicated team that manually checks Gemini's answers
Show answer and explanation

Answer: C An AI computing platform integrating TPUs, GPUs, storage and more

Infrastructure optimized for AI that raises training and inference performance and efficiency.

Q2. What is the most appropriate advantage of Google Cloud's “open approach”?

  1. It raises transparency by publishing all customers' data
  2. Using open-source models makes security management unnecessary
  3. You can also choose open and partner models, reducing lock-in to a specific technology
  4. Only Google's models can be used, so there's no confusion in choosing
Show answer and explanation

Answer: C You can also choose open and partner models, reducing lock-in to a specific technology

Model Garden, the open Gemma models, integration with other clouds and on-premises, and so on.

Reference links

2.2Ready-to-use generative AI

Ready-to-use generative AI (Gemini app and Gems, Gemini Enterprise, Gemini for Google Workspace)

Concept

The Gemini app is a conversational AI assistant, and Gems are custom Geminis that save frequently used instructions and roles. Gemini Enterprise is an enterprise platform where employees use agents connected to internal data and business apps. Gemini for Google Workspace helps with writing, summarizing and analysis inside Gmail, Docs, Sheets, Meet and more.

How it works / steps

  1. Use Gemini for Google Workspace to speed up daily document work
  2. Save repeated requests as Gems and share them
  3. Use Gemini Enterprise for search and agents across multiple internal data sources

Workplace example

A sales team shares a Gem that builds proposal outlines and has Gemini draft inside Google Docs.

Common confusions and exam tips

Mini quiz

Q1. Which is the best example of using Gems?

  1. Installing Gemini on an internal server and using it cut off from the internet
  2. Rewriting Gemini's model weights to create a company-only model
  3. Using Gems to automatically delete employee attendance records
  4. Creating a “drafter” that saves instructions for your house style and reusing it
Show answer and explanation

Answer: D Creating a “drafter” that saves instructions for your house style and reusing it

Gems are purpose-specific custom Geminis that save instructions (and reference materials).

Q2. What is the most appropriate business value of Gemini Enterprise?

  1. A feature that only auto-organizes photos on employees' personal phones
  2. A feature that only strengthens Gmail's spam filter
  3. Searching across internal data and automating work with agents
  4. A feature for measuring data center power usage
Show answer and explanation

Answer: C Searching across internal data and automating work with agents

Gemini Enterprise is a unified search, chat and agent platform for employees.

Reference links

2.3Improving customer experience

Improving customer experience (Agent Search and Google Search; Customer Engagement Suite: Conversational Agents, Agent Assist, Conversational Insights, Contact Center as a Service)

Concept

Agent Search (formerly Vertex AI Search) uses Google Search technology to build high-quality search and generated answers over your own website and documents. Customer Engagement Suite is a family of customer service products including Conversational Agents (chat and voice agents), Agent Assist (real-time support for operators), Conversational Insights (conversation analysis) and Contact Center as a Service (a cloud contact center).

How it works / steps

  1. For customer self-service, use site search and Conversational Agents
  2. For operator quality and speed, use Agent Assist
  3. For analyzing conversation trends, use Conversational Insights

Workplace example

In a call center, Conversational Agents answer common questions, and complex inquiries go to operators while Agent Assist shows answer suggestions.

Common confusions and exam tips

Mini quiz

Q1. Which is the best example of using Gemini for Google Workspace?

  1. Automatically changing the database design of core systems outside Workspace
  2. Taking Meet meeting notes and drafting a proposal in Docs
  3. Moving contact center phone lines to the cloud
  4. Training a model independently on TPUs
Show answer and explanation

Answer: B Taking Meet meeting notes and drafting a proposal in Docs

Supports creating, summarizing and organizing inside each Workspace app.

Q2. In a contact center, new agents' answer quality varies and training takes long. You want to keep human agents. Which Customer Engagement Suite capability fits best?

  1. Conversational Agents (virtual agents talk to customers directly)
  2. Contact Center as a Service (provides the contact center platform)
  3. Conversational Insights (analyzes past conversations for trends)
  4. Agent Assist (suggests answers and information to agents during calls)
Show answer and explanation

Answer: D Agent Assist (suggests answers and information to agents during calls)

Agent Assist supports human operators; it doesn't replace them. [Other options] Conversational Agents answer without people, which conflicts with keeping human agents. Conversational Insights is after-the-fact analysis and CCaaS is the platform; neither supports agents during calls.

Reference links

2.4Developer services

For developers (Agent Platform: Model Garden, Agent Search, AutoML; RAG services; building custom agents)

Concept

Agent Platform (formerly Vertex AI) is a developer platform bringing together using, tuning, evaluating and deploying models and building agents. Model Garden is a catalog of Google, partner and open models; AutoML builds models from your own data with almost no code. Use RAG services to create answers based on internal data, and build and run custom agents.

How it works / steps

  1. Compare in Model Garden whether existing models are enough
  2. Consider AutoML to build prediction models from your own data
  3. Build answers based on internal data with RAG services and agents

Workplace example

An insurer builds an agent on Agent Platform that answers by referencing policy terms via RAG, checks answer quality with evaluation features, then deploys.

Common confusions and exam tips

Mini quiz

Q1. What is the most appropriate advantage of using Model Garden?

  1. All non-Google models are prohibited
  2. All models are free with no usage limits
  3. An optimal business app is completed automatically without choosing a model
  4. You can compare many models, choose one that fits, and try it right away
Show answer and explanation

Answer: D You can compare many models, choose one that fits, and try it right away

You can choose Gemini, Gemma, Llama, Claude and others.

Q2. Which situation suits AutoML best?

  1. You have few experts but want to build a prediction model from your own labeled data
  2. You want text answers about the latest news based on web information
  3. You want to generate video ads from text instructions
  4. You want to auto-summarize employees' meeting notes inside Workspace
Show answer and explanation

Answer: A You have few experts but want to build a prediction model from your own labeled data

AutoML automatically builds custom models from labeled data.

Reference links

2.5Agent tools

Agent tools (extensions, functions, data stores, plugins; Cloud Storage, Cloud Run and AI APIs) and choosing between Agent Studio and Google AI Studio

Concept

Agents use tools to access outside information and systems. Tools include extensions (call external APIs), functions (calls to processing run on the app side), data stores (data to search) and plugins, combined with Cloud Storage, Cloud Run and various AI APIs. Agent Studio is an environment for building and managing agents for enterprises; Google AI Studio is an environment for easily trying the Gemini API and building prototypes.

How it works / steps

  1. Sort out the information (data stores) and operations (functions, extensions) the agent needs
  2. Use Google AI Studio for quick prototypes and Agent Platform for enterprise production
  3. Minimize tool permissions

Workplace example

A developer tries prompts in Google AI Studio, and in production adds the internal inventory API as a function to an Agent Platform agent.

Common confusions and exam tips

Mini quiz

Q1. A business team wants its agent to read invoices and translate them. Where should the development team go to find and enable the Google pretrained APIs the project can use?

  1. Google Cloud API Library
  2. Agent Studio
  3. Google AI Studio
  4. Model Garden
Show answer and explanation

Answer: A Google Cloud API Library

Find and enable APIs that can be used as agent tools. [Other options] Model Garden is a catalog of foundation models, Agent Studio is an environment for building agents, and Google AI Studio is for quickly trying the Gemini API; none is where you enable APIs.

Q2. What is the most appropriate value of low-code/no-code AI tools?

  1. All code written by developers becomes unusable
  2. AI quality evaluation and security management all become unnecessary
  3. AI usage fees always become free
  4. Business staff can build AI apps with almost no code
Show answer and explanation

Answer: D Business staff can build AI apps with almost no code

Democratizing AI. Governance mechanisms are still needed.

Reference links

Lessons are written by this site based on public information and do not guarantee exam content. Product specifications change, so check the latest information at the reference links.

Notes (unofficial, original questions, no pass guarantee)