The area that changed most in the August 2026 revision of the Cloud Digital Leader exam guide is AI. Domain 3 (about 18%) now includes agentic AI, Gemini Enterprise Agent Platform, and AI Hypercomputer, and Domain 1 also asks you to define agentic AI and recognize AI Hypercomputer as one of Google Cloud's differentiators. On top of that, Vertex AI was renamed Gemini Enterprise Agent Platform in April 2026, so if you study from older material you may not recognize the product names that appear in the answer choices.
This post starts with the difference between generative AI and agentic AI, then walks through how agents change the way people work, the strengths of Google Cloud AI, how Agent Platform is organized, the criteria for choosing the right AI solution, AI Hypercomputer, and data quality. The basics of AI, ML, and deep learning, along with BigQuery ML and responsible AI, are covered first in AI/ML Innovation Services.
---
Generative AI and Agentic AI: AI That Answers vs. AI That Works
If generative AI is an assistant that answers your questions and writes first drafts, agentic AI is an employee who, given the goal "get me ready for next week's business trip," breaks it into steps on their own and finishes the job: checking the calendar, searching flights, booking, and reporting back. The exam checks whether you can tell these terms apart precisely.
| Term | Meaning | Example | |------|------|------| | Artificial intelligence (AI) | The broad field of building systems that learn, reason, and make judgments like people | Chatbots, image recognition | | Machine learning (ML) | A way of doing AI that learns patterns from data instead of hand-written rules | Demand forecasting, fraud detection | | Generative AI | AI that uses learned patterns to create new content such as text, images, code, or audio | Drafting marketing copy, summarizing documents | | Agentic AI | AI that takes a goal, plans on its own, and uses tools to carry a multi-step task through to the end | Processing customer refunds, automating report writing | | Data analytics | Examining data to find out what happened and why | Finding the cause of last quarter's sales drop | | Business intelligence (BI) | The tools and processes that turn analysis into dashboards and reports for decision-making | A sales dashboard for executives |
Agentic AI usually repeats the following four steps until it reaches the goal.
Understand the goal: figure out the result the user wants Plan: break the goal into smaller steps Use tools: take actions such as searching, querying a database, or calling another system's API Check and adjust: review the result and, if it falls short, revise the plan and try again
Exam tip: "It plans on its own and moves across multiple systems to finish the work without a person directing each step" describes agentic AI. If it stops at "creates a new sentence or image," that is generative AI.
---
How Agentic AI Changes Work
The exam guide lists six areas where agentic AI is reshaping industries and the way people work. On the exam this shows up as choosing "the value an AI agent brings in this situation."
| Area | What the agent does | Business value | |------|------|------| | Workforce productivity | Moves across email, documents, and internal systems to find information, draft content, and follow up after meetings | Less time on repetitive work, so people focus on judgment calls | | Customer support | Understands what the customer is asking, looks up orders, and carries refunds and exchanges through | Instant 24/7 responses, with human agents freed for complex cases | | Sales experiences | Gathers customer history and competitor information to prepare meetings and write tailored proposals | More personalized proposals and a shorter sales cycle | | Product innovation | Explores ideas, generates prototype code and designs, and analyzes user feedback | Faster time to market | | Operations | Watches inventory, logistics, and IT alerts and takes predefined actions | Faster incident response and processing | | Research | Searches and summarizes huge volumes of documents and organizes hypotheses | Less research time and faster discovery |
Business Problems ML Is Especially Good At
Even without agents, there are three classic problems where ML on its own creates value.
Replacing or simplifying systems where people keep adding rules forever (for example, spam filters and fraud detection) Deriving insights people would miss from large volumes of structured and unstructured data (for example, analyzing hundreds of thousands of customer reviews) Scaling decisions that people used to make one at a time (for example, product recommendations and loan review support)
Exam tip: "There are so many exception rules that the system is hard to maintain" is a signal to replace the rule-based system with ML.
---
Five Strengths of Google Cloud AI
The exam guide sums up the key benefits of Google Cloud's AI offerings in five points. They overlap with "Google Cloud's differentiators" in Domain 1, so memorizing them together saves effort.
| Strength | Meaning | Representative products | |------|------|------| | Best infrastructure for AI | Hardware, software, and pricing optimized together for training and inference | AI Hypercomputer, TPUs, GPUs | | AI-ready data cloud | Analyze scattered data in one place and connect it straight to AI | BigQuery, AlloyDB, Cloud Storage | | Sophisticated first-party (1P) models | State-of-the-art models built by Google itself | Gemini, Veo | | All-in-one AI developer platform | From choosing a model to building, deploying, and managing agents, all in one place | Gemini Enterprise Agent Platform | | Pre-built AI agents and applications | Agents and applications you use right away without building them | Gemini Enterprise, Agent Garden |
First-party (1P) means a model Google built itself. That does not mean you can use only Google's models. Model Garden lets you choose from more than 200 models, including other companies' models and open models alongside 1P models such as Gemini, which gives you choice and flexibility instead of locking you into one vendor's models.
---
Gemini Enterprise Agent Platform: Formerly Vertex AI
On April 22, 2026, Google renamed Vertex AI to Gemini Enterprise Agent Platform (Agent Platform for short). The official documentation describes it as "an evolution of Vertex AI" that supports the complete AI lifecycle, from accessing foundation models to deploying and managing agents. If older material or a question mentions Vertex AI, read it as the same platform.
The Renamed Products at a Glance
| Former name | New name | |------|------| | Vertex AI | Gemini Enterprise Agent Platform | | Vertex AI Studio | Agent Studio | | Vertex AI API | Agent Platform API | | Vertex AI Model Garden | Model Garden (name kept) | | Vertex AI AutoML | Agent Platform AutoML | | Vertex AI Agent Engine | Agent Runtime | | Vertex AI Search | Agent Search |
Four Pillars: Build, Scale, Govern, Optimize
Agent Platform is organized around four pillars. The key point is that it does not stop at building agents: running them in production, controlling their permissions and security, and continuously improving their quality all happen on the same platform.
| Pillar | What it does | Representative components | |------|------|------| | Build | Design and develop agents and models | Agent Studio (low-code), Agent Development Kit (code), Agent Garden (pre-built agents and templates), Model Garden | | Scale | Run the agents you built in production | Agent Runtime, Sessions and Memory Bank (conversation context and long-term memory) | | Govern | Manage agent identity, permissions, and security policies | Agent Registry, Agent Identity, Agent Gateway, Model Armor | | Optimize | Evaluate, monitor, and improve quality | Agent evaluation, simulation testing, observability (traces and logs) |
!The four pillars of Gemini Enterprise Agent Platform and the AI Hypercomputer underneath them
Remember that the path also depends on the builder's skill level. People starting without code use Agent Studio, and developers building sophisticated agents in code use the Agent Development Kit (ADK). If you want to train a custom model on your own data while keeping code to a minimum, you use AutoML. That is why the exam guide lists Agent Studio and AutoML together as examples of "how an organization can build custom models using its own data to create business value."
Exam tip: "We want to compare models from several companies and pick one" points to Model Garden. "Design prompts and agents without code" points to Agent Studio. "Get started quickly from ready-made agent templates" points to Agent Garden.
---
Choosing the Right AI Solution
The exam guide lists five strategic considerations organizations must weigh when selecting a Google Cloud AI solution.
Implementation speed: how quickly can we use it? Development effort: how much development does it take? Potential for business differentiation: can it create value our competitors do not have? Technical expertise requirements: do we need ML engineers? Choice and flexibility: can we swap models and tools?
Most of these trade off against each other. Ready-made solutions are fast but hard to differentiate, while models you build yourself differentiate strongly but take time and specialists.
| Option | Implementation speed | Expertise needed | Differentiation potential | When it fits | |------|------|------|------|------| | Pre-built agents and apps such as Gemini Enterprise | Fastest | Almost none | Low | You want to raise employee productivity right away | | Pre-trained APIs, Gemini API | Fast | Low (API calls) | Low to medium | Adding common capabilities such as translation, speech recognition, or image analysis to an app | | Agent Studio, AutoML | Medium | Medium (minimal code) | Medium | You need models and agents fitted to your own dat