Session 1: Foundations — Understanding and Running an AI Agent
Session Overview
| Item | Details |
|---|---|
| Duration | 120 minutes |
| Objective | Understand the difference between AI chat and AI agents, lower the psychological barrier to using the terminal, and get Claude Code running on your own PC |
| Prerequisites | Completed Session 0, bring a Mac/Windows/Linux PC, internet connection |
| Participant Goals | Install Claude Code, issue your first instruction, and verify the output. Understand the role of CLAUDE.md |
Pre-Session Preparation (For Instructors)
Required Environment
-
Verify participant PC requirements in advance
- Node.js v18 or higher must be installed
- Terminal (Mac: Terminal.app, Windows: PowerShell or Windows Terminal) must be available
- Prepare Anthropic API keys for all participants or ask participants to obtain them in advance
-
API Key Preparation (Important)
- Option A: Temporarily distribute organization API keys to participants (revoke after the workshop)
- Option B: Ask participants to create an Anthropic account and obtain an API key in advance
- In either case, communicate key handling precautions (do not share with others, do not post publicly)
-
Pre-Setup Email (Send one week before Session 1)
Subject: [Session 1 Preparation] Please Install Node.js In our next Session 1, you will be running an AI agent on your own PC. Please install Node.js in advance. ▼ Installation Steps 1. Go to https://nodejs.org/ 2. Click the button labeled "LTS" to download 3. Open the downloaded file and follow the on-screen instructions to install ▼ Verifying the Installation - Mac: Open the "Terminal" app, type node -v and press Enter - Windows: Open "PowerShell", type node -v and press Enter - If a version number (e.g., v20.11.0) is displayed, you're all set -
Materials to distribute on the day
- participant-handout.md (printed or digital)
- API keys (individually printed on paper, or sent via chat individually)
Rehearsal Checklist
- Re-verify the Node.js installation steps on your own PC
- Confirm that
npm install -g @anthropic-ai/claude-codecompletes successfully - Confirm that the
claudecommand launches - Run a simple research task with Claude Code and measure the time it takes
- Rehearse the CLAUDE.md creation demo once
- Confirm that the Wi-Fi connection is stable
- Verify projector/screen sharing functionality
- Check how the screens look on both Windows and Mac (prepare both if possible)
Timetable
1. Review of the Previous Session and Homework Sharing (10 minutes)
Sample instructor dialogue:
“In the previous Session 0, you experienced how writing a request in a GitHub Issue caused an AI agent to automatically research and create a report for you. I think many of you thought, ‘That’s amazing!’ For your homework, you were asked to think of tasks you could delegate to an AI agent — would anyone like to share?”
What to do:
- Have 2-3 participants share their homework (volunteers only, no cold-calling)
- Note the task examples on a whiteboard or in chat
- Declare today’s goal: “Today, we’ll understand how that AI agent works and learn to run it on your own PC”
Common stumbling blocks and remedies:
- If many participants didn’t do the homework → Don’t criticize; ask them to spend 30 seconds thinking of ideas on the spot
- If someone says “I can’t think of any tasks” → Offer examples like “writing a daily report,” “summarizing meeting minutes,” or “drafting an email”
2. AI Chat vs. AI Agent — What’s the Difference? (15 minutes)
Sample instructor dialogue:
“First, let’s clarify the difference between the ChatGPT or Claude chat features you normally use and the ‘AI agent’ you experienced last time.
The simplest analogy is this: AI Chat = an advisor. You ask a question and it answers. But you’re the one who does the actual work. AI Agent = a team member you can delegate work to. You hand off a task, and it thinks on its own, does the work itself, and delivers a finished product.”
Content to explain on a slide or whiteboard:
┌─────────────────────────────────────────────────────────┐
│ AI Chat │
│ │
│ You: "Think of a structure for a proposal" │
│ AI: "1. Background 2. Problem 3. Proposal │
│ 4. Expected Results would be good" │
│ You: (Open Word and start writing yourself) │
│ │
├─────────────────────────────────────────────────────────┤
│ AI Agent │
│ │
│ You: "Create a proposal" │
│ AI: ① Understand the task │
│ ② Gather necessary information │
│ ③ Plan the structure │
│ ④ Create the file │
│ ⑤ "Done — please review" │
│ │
└─────────────────────────────────────────────────────────┘
Explain the 3 key characteristics of an agent:
- Autonomy — It can make its own decisions and proceed without asking a human at every step
- Tool use — It can use “tools” such as creating files, searching the web, and executing commands
- Multi-step execution — For a single request, it can automatically execute multiple steps in sequence
Sample instructor dialogue:
“Think back to your experience last time. All you wrote in the Issue was a request like ‘Research this topic.’ The AI agent received that, did the research on its own, created the file on its own, and submitted a Pull Request saying ‘Please review.’ This is exactly the combination of autonomy, tool use, and multi-step execution in action.”
Common stumbling blocks and remedies:
- “So what’s really different from chat?” → “In chat, the answer exists only within the conversation. An agent can go beyond the conversation — it can create files, browse the web, and more. That’s the key difference.”
- “Is the agent always better?” → “No. If you just want to ask a quick question, chat is easier. Agents shine when you want to delegate a substantial piece of work.”
3. Terminal Basics — Not Scary, Just Controlling Your PC with Text (15 minutes)
Sample instructor dialogue:
“From here on, we’ll use your own PCs. First, you’ll open something called a ‘terminal,’ but I want to tell you one thing first.
The terminal is not scary.
Normally, you double-click a folder with your mouse to open it, right? The terminal just does that with text (typed characters). It’s doing exactly the same thing. It just looks different.”
Step 1: Open the Terminal (5 minutes)
The instructor screen-shares and works through it with participants.
- Mac: “Open Spotlight (⌘ + Space), type ‘Terminal’, and press Enter”
- Windows: “Search for ‘PowerShell’ in the Start menu and open it”
Sample instructor dialogue:
“Did you get a black screen (or white screen)? That’s the terminal. It looks like a hacker’s screen from a movie, but what we’ll do is simple.”
Step 2: Try 3 Commands (10 minutes)
Go through each one: instructor demos → participants execute → verify results.
Command 1: pwd (Check where you are)
“Your computer has many folders. The command to check which folder you’re currently in is
pwd. It stands for ‘Print Working Directory.’ Type it and press Enter.”
$ pwd
/Users/tanaka
“Did you see something like
/Users/tanaka? This means ‘You are currently in tanaka’s home folder.’”
Command 2: ls (List files in the current location)
“Next, let’s look at the contents of the folder you’re in. Type
lsand press Enter. It stands for ‘List.’”
$ ls
Desktop Documents Downloads Music Pictures
“You see familiar folder names, right? These are the same things you see in Finder or File Explorer.”
Command 3: cd (Move to a folder)
“Finally, let’s move into a folder. Type
cd Desktopand press Enter. It stands for ‘Change Directory.’”
$ cd Desktop
$ pwd
/Users/tanaka/Desktop
“If you check with
pwdagain, you’ll see that your location has changed to Desktop. You just did the same thing as double-clicking to open a folder, but with text.”
Also explain how to go back:
“To go back up one folder, type
cd ..— that’s two dots.”
Sample instructor dialogue (summary):
“The commands you need to remember today are just these three.
pwd= Where am I?,ls= What’s here?,cd= Move. That’s all there is to basic terminal navigation.And one more important thing: Even if you type a wrong command in the terminal, your PC won’t break. If you type a command that doesn’t exist, it’ll just say ‘command not found.’ Feel free to experiment.”
Common stumbling blocks and remedies:
| Problem | Remedy |
|---|---|
| ”Terminal” can’t be found on Mac | Guide them to Finder → Applications → Utilities → Terminal |
| Garbled characters on Windows | Change the PowerShell font settings, or suggest installing Windows Terminal |
ls shows nothing | The folder may be empty. Move to another folder like cd Desktop and try again |
| Typed the command wrong | Tell them “‘command not found’ is fine. Just type it correctly again” |
| Terminal font is too small | Point out that Ctrl/Cmd + ”+” can zoom in |
4. Installing and First Launch of Claude Code (20 minutes)
Sample instructor dialogue:
“Now it’s time to install the AI agent ‘Claude Code’ on your PC. Type the command I’m about to say into the terminal.”
Step 1: Verify Node.js (2 minutes)
$ node -v
v20.11.0
“If a version number appears, you’re good. If you see ‘command not found,’ please raise your hand. We’ll install it together.”
Handling participants who don’t have Node.js:
- The expectation is that they installed it from the advance instructions, but plan for day-of support
- Download and install the LTS version from https://nodejs.org/
- After installation, the terminal needs to be closed and reopened
Step 2: Install Claude Code (5 minutes)
$ npm install -g @anthropic-ai/claude-code
“
npmis like an ‘app store’ for Node.js. This command downloads and installs the Claude Code app from the internet. It takes a little while, so let’s wait.”
Sample instructor dialogue (while waiting for installation):
“You’ll see lots of text scrolling on the screen — this is the installation progress. As long as you don’t see errors in red text, everything is fine.”
Step 3: Set the API Key (5 minutes)
$ export ANTHROPIC_API_KEY=sk-ant-xxxxxxxxxxxxxxxx
“An API key is like a password that lets Claude Code prove to the server that it’s an authorized user. Please enter the key I’m about to distribute.
Important: Do not share this key with anyone or post it on social media. It’s as sensitive as a credit card number.”
- Distribute the API key to participants (on paper, via chat, or temporarily displayed on a slide)
- Note that on Windows the command is
$env:ANTHROPIC_API_KEY="sk-ant-xxxxxxxxxxxxxxxx"
Step 4: Launch Claude Code (5 minutes)
$ claude
“Type
claudeand press Enter to start Claude Code.”
What the screen looks like on successful launch:
╭──────────────────────────────────────────╮
│ ✻ Welcome to Claude Code! │
│ │
│ /help for help │
│ │
╰──────────────────────────────────────────╯
>
“When you see this
>prompt, you’re ready to give instructions to Claude Code. Congratulations! The AI agent is now running on your PC.”
Step 5: Your First Interaction (3 minutes)
“Try saying something to it. For example, type this:”
> Hello. What can you do?
“Did you get a response? This is still just the ‘chat’ stage. In the next section, we’ll experience the true power of the agent.”
Common stumbling blocks and remedies:
| Problem | Remedy |
|---|---|
npm: command not found | Node.js is not installed. Follow the advance setup instructions to reinstall |
| Permission error during npm install (Mac) | Re-run with sudo npm install -g @anthropic-ai/claude-code. Explain that a password is required |
| Permission error during npm install (Windows) | Re-run PowerShell with “Run as administrator” |
| API key error | Check for extra spaces before or after the key. Try wrapping it in quotes |
claude: command not found | Close and reopen the terminal. Or run with full path: npx @anthropic-ai/claude-code |
| Authentication error after launch | Verify the API key is set correctly with echo $ANTHROPIC_API_KEY |
5. Your First Task (30 minutes)
Sample instructor dialogue:
“Here’s where the real action begins. Let’s give Claude Code some ‘work’ to do. In Session 0 we did this on GitHub, but this time we’ll issue instructions directly to the AI agent right on your own PC.”
Step 1: Prepare a Working Folder (5 minutes)
Exit Claude Code with /exit, then create a working folder.
$ cd ~/Desktop
$ mkdir my-first-project
$ cd my-first-project
“We’ve created a folder called
my-first-projecton the Desktop.mkdirstands for ‘Make Directory’ — it creates a folder. Now let’s start Claude Code inside this folder.”
$ claude
Step 2: Assign a Research Task (10 minutes)
“Now let’s ask Claude Code to do some research. Use one of the examples below as a reference, and try making a request related to your own work.”
Task Example A:
> Research the top 5 trends in the SaaS industry for 2025 and compile them
in a file called report.md. Include an overview and specific examples
for each trend.
Task Example B:
> Create a comparison table of the pros and cons of remote work vs.
office work and save it in a file called comparison.md.
Task Example C:
> Create a training plan for a business etiquette workshop for new
employees and save it in a file called training-plan.md.
Assume a 60-minute session.
Step 3: Experience and Explain Permission Prompts (5 minutes)
“As Claude Code works, you may see confirmation messages like ‘May I create a file?’ This is called a permission prompt — it’s a safety mechanism to prevent the AI agent from performing dangerous operations without your consent.”
Claude wants to create file: report.md
Allow? (y/n)
“Type
y(Yes) and press Enter to continue. This is a critically important design principle. AI agents are powerful, but the final decision is always made by a human. It will always ask, ‘Is this okay?’”
Step 4: Review the Output (5 minutes)
$ ls
report.md
“If you check with
ls, you’ll see the file has been created. Open themy-first-projectfolder in Finder or File Explorer and verify that the file actually exists.Double-click it and read the contents. This is a real file created on your PC. In Session 0 it was on GitHub, but this time it happened right on your own machine.”
Step 5: Try Giving Follow-Up Instructions (5 minutes)
“If you see something you’d like changed in the file, you can give Claude Code additional instructions.”
> Please add more specific data about the 3rd trend in report.md.
“This way, you can refine the output through back-and-forth dialogue. It’s the same thing as commenting on a PR in Session 0, but done in the terminal.”
Sample instructor dialogue (summary):
“Let’s summarize what we just did.
- Created a folder
- Launched Claude Code
- Assigned a task in plain English
- The AI agent created a file
- Reviewed it and gave follow-up instructions
What we did on the GitHub interface in Session 0, we’ve now done directly in the terminal on our own PC.”
Common stumbling blocks and remedies:
| Problem | Remedy |
|---|---|
| Claude Code response is slow | Explain: “The AI is thinking. More complex tasks take longer” |
| Response is in a different language than expected | Add “Please respond in English” as a follow-up, or include “in English” in the original instruction |
| File was not created | Check Claude Code’s output. If there’s an error, ask participants to consult the instructor |
| Unsure what to choose at the permission prompt | Say “For file creation and reading, it’s generally safe to choose Yes” |
| Participant can’t think of a task | Offer the task examples and let them use one as-is |
6. Introducing CLAUDE.md (15 minutes)
Sample instructor dialogue:
“Finally, let me introduce a feature that helps you use Claude Code even more effectively. It’s a file called CLAUDE.md.
When a new member joins your team at work, you give them a ‘handover document’ or ‘operations manual,’ right? CLAUDE.md is a handover document for the AI agent.”
Step 1: The Concept of a Project Folder (3 minutes)
“Claude Code recognizes the folder where it was launched as its ‘workspace.’ It can read and write files within that folder. We call this the ‘project folder.’
The
my-first-projectfolder you created earlier is exactly that — a project folder.”
my-first-project/ ← Project folder (the AI's workspace)
├── CLAUDE.md ← Handover document for the AI
├── report.md ← Output created by the AI
└── (other files)
Step 2: Explain the Role of CLAUDE.md (5 minutes)
“If you write instructions in CLAUDE.md, Claude Code will automatically read them every time. This means you no longer need to repeat the same instructions each time.
For example, you can write things like this:”
# Project Settings
## Your Role
You are a research assistant for the marketing department.
## Working Rules
- Always create deliverables in English
- Use Markdown (.md) file format
- Always cite sources for research content
- Use tables and lists to make content readable
## Target Audience
Marketing department managers (non-technical)
“If you set this up, you won’t need to say ‘write in English’ or ‘include sources’ every time — it will follow these rules from the start.
The repository we used in Session 0 also had a CLAUDE.md file. That’s why the AI agent was able to produce deliverables in the right format.”
Step 3: Write a CLAUDE.md Together (7 minutes)
“Now let’s create a CLAUDE.md in the
my-first-projectfolder. We’ll ask Claude Code to create it for us.”
If Claude Code is not running, restart it:
$ cd ~/Desktop/my-first-project
$ claude
Give Claude Code the instruction:
> Please create a CLAUDE.md for this project.
Include the following:
- Your role: Business research assistant
- Create deliverables in English
- File format: Markdown
- Always cite sources in research
“CLAUDE.md has been created. From now on, whenever you start Claude Code in this folder, these settings will be loaded automatically.
Write the handover document once, and the AI agent will always remember it. That’s the power of CLAUDE.md.”
Sample instructor dialogue (advanced tip):
“You can freely edit CLAUDE.md. If the project changes, just update the handover document. In future sessions, we’ll learn more practical ways to write CLAUDE.md.”
Common stumbling blocks and remedies:
| Problem | Remedy |
|---|---|
| CLAUDE.md name is wrong (claude.md, Claude.md, etc.) | Explain that the uppercase name matters. Since Claude Code can create it automatically, there’s no need to create it manually |
| Don’t know what to write | Say “Start with just three things: role, language, and format — that’s enough” |
| Don’t understand the concept of a project folder | Point concretely: “The new folder you created on the Desktop = the project folder” |
7. Review and Q&A (15 minutes)
Discussion prompts (5 minutes):
- Comprehension check: “Can anyone explain the difference between AI chat and AI agent in your own words?”
- Impressions: “What was your impression of using the terminal for the first time? Was it less scary than you expected?”
- Surprises: “What left the biggest impression when you actually ran Claude Code?”
Points for the instructor to summarize (5 minutes):
“Let’s review what we learned today.”
-
AI chat and AI agents are different things
- Chat = conversation partner (answers you, but doesn’t do the work)
- Agent = team member (thinks, does the work, and delivers output)
-
The terminal is just text-based operation
pwd= Where am I? /ls= What’s here? /cd= Move- Nothing breaks if you make a mistake
-
Claude Code is an AI agent that runs on your own PC
- The same thing as the GitHub experience in Session 0, executed directly on your local PC
- Permission prompts prevent it from doing anything dangerous on its own
-
CLAUDE.md is a handover document for the AI
- Place it in the project folder and it’s automatically read every time
- Eliminates the need to repeat the same instructions
Q&A (5 minutes):
Common questions and sample answers:
| Question | Answer |
|---|---|
| ”Does the API key cost money?" | "Yes, costs are incurred based on AI usage. For today’s exercises, it’s roughly a few cents to a dollar. During the workshop, we’re using the organization’s key, so there’s no personal cost" |
| "Do I need to keep Claude Code running all the time?" | "No, you launch it when you need it and exit with /exit when you’re done" |
| "Where are the created files saved?" | "They’re saved in the folder where you launched Claude Code (the project folder). They’re regular files, so you can open them from Finder or File Explorer" |
| "Can I use it offline?" | "No, Claude Code communicates with AI servers over the internet, so an online connection is required" |
| "Is it secure?" | "Claude Code asks for confirmation before file operations. Also, as long as you manage your API key properly and avoid including confidential information in prompts, it can be used safely” |
Next Session Preview and Closing
Sample instructor dialogue:
“Today, we opened the terminal, installed Claude Code, and reached the point of delegating work to an AI agent on your own PC for the first time.
You’re probably starting to see how the ‘amazing’ experience from Session 0 actually works behind the scenes.
In the next Session 2, we’ll learn about file management (Git) and how to use AI agents in a team setting. It’s about properly managing and accumulating the deliverables that the AI agent creates.”
Homework (optional):
- Run 2 tasks related to your own work using Claude Code
- Customize the contents of CLAUDE.md to suit your needs
- Before the next session, check the files Claude Code created in the
my-first-projectfolder on your Desktop
Appendix A: Node.js Installation Steps (Detailed)
Mac
- Go to https://nodejs.org/
- Click the “LTS” download button
- Open the downloaded
.pkgfile - Follow the installer instructions (click “Continue” and “Agree” for everything)
- Open Terminal and verify with
node -v
Windows
- Go to https://nodejs.org/
- Click the “LTS” download button
- Open the downloaded
.msifile - Follow the installer instructions (click “Next” and “I accept” for everything)
- Important: After installation, close PowerShell and reopen it
- Verify with
node -v
Appendix B: Persisting the API Key Setting (Reference)
During the workshop, we set the key each time with the export command,
but the API key setting is lost when you close the terminal.
If you want to persist it, use the following methods (instructors can share during follow-up):
Mac (zsh)
echo 'export ANTHROPIC_API_KEY=sk-ant-xxxxxxxx' >> ~/.zshrc
source ~/.zshrc
Windows (PowerShell)
[Environment]::SetEnvironmentVariable("ANTHROPIC_API_KEY", "sk-ant-xxxxxxxx", "User")
Note: Only share this setup for participants who want to continue using it for self-study. Manual setup each time is sufficient during the workshop.
Appendix C: Troubleshooting Reference
| Problem | Cause | Remedy |
|---|---|---|
Error on node -v | Node.js not installed | Install using the steps in Appendix A |
Permission error on npm install -g | Insufficient permissions | Mac: add sudo at the beginning / Windows: run as administrator |
claude gives “command not found” | Path not configured | Restart the terminal, or use npx @anthropic-ai/claude-code |
| API key error | Key setting mistake | Check the set value with echo $ANTHROPIC_API_KEY |
| Claude Code not responding | Network issue | Check Wi-Fi connection, try a different network |
| Installation is extremely slow | Network bandwidth | Try tethering, or show a demo on the instructor’s PC |
Appendix D: Instructor Checklist (Day Of)
- Prepare printed Node.js installation instructions
- Prepare API keys for each participant (individually on paper or via chat)
- Final confirmation that Claude Code works on your own PC
- Be prepared to explain how to open the terminal on both Mac and Windows
- Verify Wi-Fi connection stability
- Verify projector/screen sharing functionality
- Prepare a backup mobile hotspot (optional)
- Prepare participant-handout.md for distribution (printed or digital)