Homework: Build Your First AI Agent | Launch With Us
Homework Assignment

Build Your First AI Agent

Complete these 4 tasks before our next session. Estimated time: 30–45 minutes. No coding experience required.

?!

Stuck at ANY Point? Screenshot It.

If you hit an error, take a screenshot and paste it into Claude (any version — web, desktop app, or terminal). It will read the error and tell you exactly how to fix it.

1
Install Claude Code

Get Claude Code installed and authenticated. Takes 5–10 minutes.

Step 1 of 5

Open your terminal

Mac

Press Cmd + Space, type Terminal, press Enter.

Windows

Press Win + R, type cmd, press Enter.

What is the terminal?

It's a text-based way to talk to your computer. Instead of clicking icons, you type commands. Think of it as texting your computer.

Step 2 of 5

Check if Node.js is installed

Type this and press Enter:

$ node --version

If you see v18 or higher, you're good. If you see "command not found," go to nodejs.org, click the green "LTS" button, install it, then close and reopen your terminal and try again.

Step 3 of 5

Install Claude Code

$ npm install -g @anthropic-ai/claude-code

Wait for it to finish (1–2 minutes). If you get a "permission denied" error, use:

$ sudo npm install -g @anthropic-ai/claude-code

It will ask for your computer login password. Characters won't appear as you type — that's normal.

Step 4 of 5

Verify it works

$ claude --version

You should see a version number. If you see "command not found," close your terminal, reopen it, and try again.

Step 5 of 5

Launch and authenticate

$ claude

The first time you launch Claude Code, it will walk you through three things:

  • Terms of Service — Type y and press Enter to accept
  • Authentication — Choose "Anthropic API Key" (paste the key your manager gave you) or "Claude App" (logs in through your browser with your subscription)
  • Permission mode — Pick "Suggest" for now (safest for beginners — more on this in Task 3)
When you're done, you'll see

Claude Code is ready! — You're in. Claude is listening.

2
Three Ways to Use Claude

Claude Code works on your terminal, desktop app, and phone. Each has strengths. Here's when to use what.

Choose Your Interface

Terminal, Desktop App, or Claude.ai — it's the same brain

No matter which way you access Claude, it's the same AI. The difference is where it runs and what it can do from there.

APP

Desktop App

A visual window (Mac/Windows). Same power as the terminal but with a friendlier interface. Great for people who prefer not to use the command line.

Best for: Everyday use once you're comfortable
WEB

Claude.ai / Phone

Access Claude via claude.ai in any browser, or through a remote session on your phone. Good for chatting, reviewing, and quick tasks on the go.

Best for: Reviewing work, quick questions, mobile access
Which should I use for this homework?

Use the terminal. It's the only way to install Claude Code and build your first agent. Once you're comfortable, you can switch to the desktop app for daily use.

Desktop App

How to get the Desktop App

Go to claude.ai/download in your browser and download the app for your platform (Mac or Windows). Once installed, sign in with the same account you used in the terminal. Everything syncs.

When to use the desktop app

Once you've completed this homework and are comfortable with Claude, the desktop app is the most convenient way to use it day-to-day. It has the same features as the terminal, but with a more visual, chat-like interface.

Claude.ai / Phone Access

How to use Claude on your phone or from anywhere

Option A — claude.ai: Open claude.ai in any browser. Log in with your Anthropic account. You can chat with Claude, but it won't have direct access to your computer's files.

Option B — Remote Session: In your terminal, run:

$ claude --remote

This prints a URL. Open that URL on your phone or any device. Now you're controlling your computer's Claude Code session from your phone — it can still read/write files on your laptop.

Important

Remote sessions only work while your terminal stays open and your computer stays awake.

3
Permissions & Control

How the agent comes alive, how you stay in control, and how to change settings. 5 minutes to read.

How it works

The agent comes alive when you give it a task

When you type a request (like "research Nike's brand strategy"), Claude starts working. Depending on your permission mode, it will either:

  • Ask before every action (Suggest mode) — you press Enter to approve each step
  • Edit files freely but ask before running commands (Auto-edit mode)
  • Do everything automatically (Full auto mode) — not recommended for beginners

The agent keeps working until the task is done or you stop it.

Approving Actions

When Claude asks for permission, here's what to press

Whenever Claude wants to create a file, edit something, or run a command, it shows you what it wants to do and asks Allow? [Y/n/?]

  • Enter or y — Yes, do it (you'll press this 95% of the time)
  • n — No. Claude will ask what you'd prefer instead
  • ? — Explain what this action does in plain English
Golden rule

When in doubt, press ? first to get an explanation, then y if it makes sense.

Stopping the Agent

How to stop Claude at any time

  • Ctrl + C — Interrupts whatever Claude is doing right now. You can then give new instructions or exit
  • /exit — Leaves Claude Code and returns to your normal terminal
  • Just close the terminal window — everything stops

Claude will never keep running after you close the terminal or press Ctrl+C. You are always in control.

Changing Permissions

How to change your permission mode anytime

From inside Claude Code (terminal):

/permissions

This re-opens the permission menu. Use arrow keys and Enter to pick a new mode.

From the Claude.ai web app:

Go to claude.ai → Settings → Claude Code. You can adjust permission defaults, review which tools Claude can access, and manage your connected apps (MCPs) from there.

From the Desktop app:

Open the app → click the gear icon (Settings). Same options as the web app — permission modes, connected tools, and preferences.

Recommended progression

Week 1: Start with Suggest mode. Approve everything manually.
Week 2: Move to Auto-edit. Claude handles files, you approve commands.
Week 3+: Try Full auto for trusted projects.

4
Build a Brand Research Agent

The main event. Create an AI agent that researches any brand on command. 20–30 minutes.

Step 1 of 5

Set up your project folder

In your terminal (outside of Claude Code — type /exit first if you're still in it):

$ mkdir ~/Desktop/my-first-agent
$ cd ~/Desktop/my-first-agent
$ git init
$ claude

This creates a folder on your Desktop, sets up Git (your safety net for undoing mistakes), and starts Claude Code.

"git: command not found"?

Mac: Type xcode-select --install, click Install, wait 5–10 min.
Windows: Download from git-scm.com, install, reopen terminal.

Step 2 of 5

Design your research workflow (pen and paper first!)

Before typing anything into Claude, think about what a great brand research report looks like to you. Jot down:

  • What sections should the report have? (overview, competitors, audience, messaging?)
  • What data matters most to YOUR role? (SEO metrics, sentiment, ad spend?)
  • How should the final output look? (executive summary, tables, bullet points?)
This is the whole point

There's no right answer. A social media manager's agent looks different from a content strategist's. Your marketing expertise shapes how the agent thinks.

Step 3 of 5

Give Claude your instructions

Type a prompt like this into Claude Code (customize it with YOUR sections):

I want to build a brand research agent. When I give it a brand name, it should:

1. Research the brand's online presence
2. Identify their top 3 competitors
3. Analyze their messaging and positioning
4. Summarize audience demographics
5. Note any recent campaigns or news

Create this as a reusable skill that outputs a formatted markdown report.

Claude will start building. It will ask permission for creating files and running commands. Read each prompt and press Enter to approve.

Step 4 of 5

Test it, then refine at least 3 times

Once Claude finishes building, test it:

Run the brand research skill on "Nike"

The first output won't be perfect. That's the point. Give Claude feedback to improve it:

# Examples of feedback to give:
The competitor section needs more detail. Add a comparison table.

Add a "Content Strategy" section that analyzes their social media cadence.

Create a one-page executive summary at the top.

# After each round, tell Claude:
Update the skill with these changes so it does this automatically next time.
Why iterate?

Each round of feedback permanently improves the skill. After 3 rounds, your agent produces a much better report — automatically, every time.

Want a dashboard instead of a report?

Just ask for one. Try: "Turn this brand research into a visual HTML dashboard with charts and key metrics." Claude will build it for you — no coding needed. You can ask for any output format: dashboards, slide decks, spreadsheets, websites. If you can describe it, Claude can build it.

Step 5 of 5

Save everything

/commit

Claude writes a commit message describing what you built. Press Enter to approve.

!
Quick Troubleshooting

Common issues and how to fix them.

"command not found: node"

Node.js isn't installed. Go to nodejs.org, click "LTS," install with defaults. Close and reopen your terminal.

"command not found: claude"

Close and reopen your terminal. If still not found, reinstall: npm install -g @anthropic-ai/claude-code. Permission error? Add sudo before the command.

"API key invalid" or authentication failed

Run claude again. Make sure you copy the entire key (starts with sk-ant-) with no extra spaces. Or try "Claude App" login instead.

Claude seems frozen (no response for 2+ minutes)

Press Ctrl + C to cancel. Try a simpler request. If it keeps freezing, type /exit, wait 10 seconds, type claude to restart.

I accidentally approved something I shouldn't have

This is why we use Git! Tell Claude: "Undo the last change" or "Revert to the previous commit." It will roll everything back.

I see an error I don't understand

Take a screenshot (Cmd+Shift+4 on Mac, Win+Shift+S on Windows). Paste it into Claude Code or claude.ai. Ask: "I got this error while trying to [what you were doing]. How do I fix it?"

- - -

What to Bring to Our Next Session

When we meet again, be ready to share:

  • Your brand research agent's output for a brand of your choice
  • Your thought process — why did you structure it the way you did?
  • What you'd change or add after seeing the results
  • One thing that surprised you about working with Claude Code
Why we're doing this

Everyone's agent will follow a different thought process. When we meet again, we'll compare all of them, find which approach works best, and combine the best pieces into one ultimate brand research agent — built by the whole team.