How to use this book
Who this book is for
This is a Green Belt course book for people who want to cut their team's recurring work without code by combining projects, custom assistants, connectors and agent features. It is written with three groups of readers in mind.
| Reader | What you will mainly gain from this book |
|---|---|
| Team leads who want to reduce recurring work with AI | Criteria for deciding what to delegate and how far, and ways to demonstrate the benefits in numbers |
| Administrators of internal AI tools | Ways to control sharing, permissions and connections across an organization, and the distinction between data retention and training use |
| People responsible for no-code automation | Ways to design instructions and reference materials, and place approval points in connector and agent tasks |
We recommend White Belt (AI Literacy) level fundamentals and at least six months of experience using AI tools as a team. We assume you know what a prompt is and that AI can produce plausible but incorrect answers. The book does not cover writing code yourself or API integration. Even topics that border on development, such as MCP, are explained only from the perspective of someone choosing and checking connections.
When you finish this book, you should be able to:
- Break work into units that can be automated, and choose the right combination of features and human review points.
- Design knowledge spaces and custom assistants and roll them out to your team.
- Grant connectors and agent features only the minimum necessary permissions.
- Prevent data leakage by configuring sharing, permission and retention settings.
- Review automation output and improve it with operational metrics.
How this book is organized
Chapters 1–5 follow the five domains of the Promotion exam in order. The output of each chapter becomes material for the next. After breaking work into steps in Chapter 1, you build the knowledge spaces and assistants for those steps in Chapter 2, connect external systems in Chapter 3, define sharing boundaries in Chapter 4, and operate and improve the automation in Chapter 5.
Every chapter has the same ten sections.
| Section | Purpose |
|---|---|
| Learning objectives | What you should be able to do after finishing the chapter |
| Introduction | A short scene showing why the chapter is needed |
| Key concepts | Topic-by-topic explanations of all knowledge and skills in the exam scope |
| Going deeper | Further discussion of trade-offs, exceptions and failure analysis |
| At work | How a fictional organization moves from a problem to a decision, including rejected options |
| Common mistakes | Plausible misconceptions and why they are wrong |
| Summary at a glance | A summary to revisit the day before the exam |
| Check the official docs | Official pages for checking details |
| Self-check | Questions to check your understanding; answers and explanations are in Appendix B |
| Hands-on tasks | Exercises that produce a concrete result in 30–60 minutes |
Appendix A is a glossary, Appendix B explains the self-check answers, Appendix C is an exam day guide, and Appendix D lists the official documents cited in this book. On your first reading, focus on Key concepts and At work in each chapter. On your second reading, study Going deeper and Common mistakes carefully.
The exam at a glance
| Item | Promotion exam | Free practice set |
|---|---|---|
| Number of questions | 30 | 10 |
| Time limit | 45 minutes | 20 minutes |
| Passing score | A raw score of 70 or more | None |
| Chapter | Domain | Weight |
|---|---|---|
| Chapter 1 | Work analysis and automation design | 20% |
| Chapter 2 | Knowledge spaces and custom assistants | 25% |
| Chapter 3 | Connectors and agent features | 25% |
| Chapter 4 | Sharing, permissions and data boundaries | 15% |
| Chapter 5 | Operations and improvement | 15% |
The raw score is the sum of the points for correctly answered questions divided by the total available points, converted to a score out of 100. Questions that test concept recognition or the application of a single principle are worth 2 points; analysis and judgment questions requiring two or more conditions to be considered together are worth 3. At least half of the Green Belt Promotion exam consists of analysis and judgment questions, so missing difficult questions costs more points. Results show whether you passed and a grade from 1 to 9. A score of 90 or more is Grade 1, 80 or more is Grade 2, and so on in ten-point intervals; all scores below 20 are Grade 9.
What the questions look like
Every question on the exam paper is answered with a single number (①–⑤). Five question formats are presented in this form.
| Format | How it appears on the exam paper | Requirement for points |
|---|---|---|
| Single choice | Choose one of five options | One correct answer |
| Multiple choice | Choose the correct combination of statements A, B, C… in the <Statements> box |
A combination containing every correct answer and no incorrect ones |
| Statement evaluation | Choose the combination of correct statements from three or four statements | A statement is correct only when both the judgment and the reason are correct |
| Ordering | Choose from five candidate orders for steps (A)–(E) | The entire order must be exact |
| Matching | Choose from five candidate mappings between items and their matches | Every match must be exact; one match may be unused |
Analysis and judgment questions include two or three conditions in the scenario, and changing just one can change the correct answer. The following example was newly written for this book to illustrate the format and is unrelated to actual exam questions.
Note
Example. An internal library administrator wants to build an assistant that answers questions about borrowing books. Borrowing rules change every quarter, and only the administrator should see the overdue borrowers list. Which design is most appropriate?
① Upload both the rules and the overdue borrowers list as knowledge, and write “Do not disclose the list” in the instructions. ② Upload the rules as knowledge and replace them whenever they are revised; leave the overdue borrowers list out of the assistant. ③ Paste the rules into the instructions field and upload the list as a separate file. ④ Include no rules, and have the administrator answer questions directly. ⑤ Combine all materials into one file to make retrieval easier.
The answer is ②. Option ① mistakes instructions for access control; ③ puts frequently changing reference material in the instructions and uploads a list meant only for the administrator; ④ abandons the purpose of automation; and ⑤ mixes sensitive material into a shared space. Only ② satisfies both conditions: “changes every quarter” and “only the administrator should see it.”
Six-week study plan
The plan assumes 40 minutes a day, four days a week. Apply the hands-on tasks to real work, but you can still achieve the learning objectives by using invented materials for steps that involve sending externally or changing real organizational settings.
| Week | Reading | Deliverable | Check |
|---|---|---|---|
| Week 1 | This guide, Chapter 1 | A workflow step table and automation design sheet for one team task | Chapter 1 self-check |
| Week 2 | Chapter 2 | Team assistant instructions, a materials register and a test question set | Chapter 2 self-check, one free practice set |
| Week 3 | Chapter 3 | A connection permission checklist and an agent task specification | Chapter 3 self-check |
| Week 4 | Chapter 4 | A sharing audit checklist and a plan to separate sensitive materials | Chapter 4 self-check |
| Week 5 | Chapter 5 | An operational metrics definition table and a failure improvement record | Chapter 5 self-check, two free practice sets |
| Week 6 | Full review, appendices | Review each chapter's Summary at a glance and reattempt incorrect self-check questions | Check Appendix C, mock exam |
Notation and sources
- Product information in this book was checked against each company's official help pages and documentation on October 20, 2026. The pages checked are collected in each chapter's Check the official docs section and in Appendix D.
- Product, feature and setting names retain their original on-screen spelling (Projects, Can view, Always ask, and so on). Their meaning is explained on first use.
- Information that could not be verified in official documentation is explained only as a general principle, rather than attributed to a particular company's features. Frequently changing figures, such as prices, model versions and detailed limits, are omitted unless essential.
- All companies and individuals in the examples are fictional. They are unrelated to real businesses or customers.
- Laws are cited only through provisions checked at Korea's National Law Information Center. The explanations in this book are not legal advice.
This field changes very quickly. The table below shows the product names corresponding to the feature types used in this book as of the verification date. The exam asks about feature types and principles rather than product names.
| Form | OpenAI | Anthropic | |
|---|---|---|---|
| Knowledge space | ChatGPT Projects | Claude Projects (new version in phased beta rollout) | Gemini Notebook (formerly NotebookLM) |
| Custom assistants and reusable instructions | GPTs (custom GPT retirement planned; migration to Plugins recommended) | Project instructions, Skills | Gems (personal accounts move to skills from November 2026) |
| External connections | Apps, Plugins bundling apps or skills | Connectors (including remote MCP servers) | Google Workspace connections (such as Drive files) |
| Multistep work and recurring runs | ChatGPT Work, Scheduled tasks, cloud browser | Claude Cowork, scheduled tasks, Claude in Chrome | Schedules (scheduled runs) and tasks in the Gemini app (personal accounts, selected plans) |
Caution
OpenAI states that the former ChatGPT agent is no longer available and directs users to ChatGPT Work for long, multistep work. Custom GPTs are scheduled to retire in Enterprise workspaces on December 11, 2026. If the product interface differs from this book when you take the exam, consult the official documentation first.