繁中Beginner guide

AI beginner guide · 9 units · 31 lessons

AI for Beginners: From Generative AI Basics to Your First Verifiable Workflow

A practical learning path for knowledge workers who want to use AI and Codex at work without skipping verification, data boundaries, or human responsibility. No programming background is required.

Author
Aaron Huang
Updated
Status
31 lessons published
Start with lesson 1 See all 9 units

Start here

Where should an AI beginner begin?

Begin with a bounded task whose result you can inspect. Do not start by automating a high-risk workflow. First learn what the model can get wrong, define the input and acceptance conditions, verify the output, and only then preserve the method as a repeatable workflow.

A usable starting method

Choose a bounded task and give AI enough context to work.

A verification habit

Check sources, numbers, data boundaries, and acceptance conditions.

A controlled workflow

Preserve permissions, human approvals, versions, and stop conditions.

Complete learning path

Nine units, in a fixed learning order

Each lesson is a published article. The course position does not change when new articles are added to the site.

Unit 1 · 4 lessons

Understand AI and Codex before starting

Understand generative AI's limits, choose the right workspace, and set a safe permission boundary. Takeaway: A capability boundary map and workspace permission checklist
  1. M1How Does Generative AI Produce Answers—and Why Can It Be Confidently Wrong?How generative AI answers
  2. M2ChatGPT Chat vs Work vs Codex: Differences and Which to UseChoose a starting workspace
  3. M3Codex for Beginners: Understand the Workspace, Conversation, File Changes, and Command ResultsOrient yourself in Codex
  4. M4How to Give Codex Full Access (and When Not To)Set permissions and scope

Unit 2 · 3 lessons

Give AI its first bounded task

Choose a low-risk, checkable first task and complete one human-reviewed delivery loop. Takeaway: A first-task brief and human acceptance record
  1. M1How to Choose Your First AI Task: Start Small, Low-Risk, and ReviewableChoose your first AI task
  2. M2How to Complete Your First Reviewable Task With CodexComplete your first Codex task
  3. M3AI Meeting Notes: Turn Source Text into Decisions and Action Items You Can ReviewProduce and revise

Unit 3 · 3 lessons

Describe and revise the work

State the goal, context, examples, and output format, then revise the first draft precisely. Takeaway: A reusable task instruction and revision log
  1. M1How to Write Better AI Prompts: Define the Work Before You Tune the PromptFrame the task
  2. M2Should You Add Context, Examples, or a Reference Format?Add context, examples and formats
  3. M3How to Fix an AI First Draft Without Starting OverRevise the first draft

Unit 4 · 3 lessons

Decide whether an AI result is usable

Verify sources and numbers while separating facts, inferences, and human-owned decisions. Takeaway: An answer checklist and fact/inference/decision labels
  1. M1Can You Trust an AI Answer? Five Checks Before You Use ItVerify the answer
  2. M2How to Check Whether AI Citations and Numbers Are RealVerify citations, numbers and sources
  3. M3How Should You Verify AI Facts, Inferences, Advice, and Decisions?Separate facts, inferences and decisions

Unit 5 · 3 lessons

Data, permissions, and human responsibility

Decide which data is usable, who has authority, and when to de-identify or synthesize it. Takeaway: A data classification and authority matrix
  1. M1Can You Put Company Data into AI? Four Checks Before You Upload ItCheck data and authority
  2. M2How ChatGPT Data Access Changes Across Personal, Managed Accounts, and AppsCheck account and app data boundaries
  3. M3How to De-Identify Data or Build Synthetic Practice MaterialDe-identify or synthesize practice data

Unit 6 · 4 lessons

Give AI durable work context

Use project structure, instructions, version records, and a Memory Router to reduce repeated explanation, false retrieval, and context drift. Takeaway: A project context, instruction, version, and memory-routing structure
  1. M1Should AI Work Live in a Project, Folder, or Conversation?Organize projects, folders and chats
  2. M2Which Instructions Belong in AGENTS.md?Save instructions and AGENTS.md
  3. M3Why Does AI Forget, and How Should You Manage Versions?Manage context, memory and versions
  4. M4How to Build AI Long-Term Memory with a Memory RouterBuild long-term memory and retrieval rules

Unit 7 · 3 lessons

Turn useful methods into repeatable workflows

Identify repeatable methods and decide whether they belong in a Skill, Plugin, Connector, or MCP. Takeaway: A reusable workflow spec and implementation choice
  1. M1A Repeatable AI Image Workflow: From Brief to Reviewed PromptBuild a repeatable image workflow
  2. M2What Is a Skill, and Which Work Is Worth Turning Into One?Identify a Skill candidate
  3. M3What Problems Do Plugins, Connectors, and MCP Solve?Compare Plugins, Connectors and MCP

Unit 8 · 4 lessons

Move from chat to controlled agent work

Run an Agent experiment with explicit permissions, approvals, and stop conditions before automating it. Takeaway: An Agent experiment plan and control boundary
  1. M1What Is an AI Agent? Chat, Work, and Agentic Tasks ExplainedCompare Chat, Work and Agent
  2. M2How to Run a Safe First AI Agent ExperimentRun the first Agent experiment
  3. M3How to Design Agent Permissions, Approvals, Stops, and RecoverySet permissions, approvals and stops
  4. M4How to Prove an AI Workflow Is Repeatable Before AutomationProve the workflow before automating

Unit 9 · 4 lessons

Move from individual use to team adoption

Turn an individual method into a small team pilot that is executable, testable, and owned. Takeaway: A team SOP, reliability test, and responsibility matrix
  1. M1How to Build a Team SOP From a Personal AI WorkflowWrite the team AI SOP
  2. M2How to Test Whether a Skill, Plugin, or Workflow Is ReliableTest workflow reliability
  3. M3How Should Teams Assign Responsibility for AI Work?Assign data, version and approval ownership
  4. M4How to Design a Small, Measurable, Stoppable AI PilotDesign the first team pilot

If you already use AI at work

Choose the unit that matches your current bottleneck.

Use Units 1–3 for task framing, Units 4–5 for verification and data boundaries, Units 6–8 for repeatable and agentic work, and Unit 9 for team adoption.

Browse all articles