Foundations Module 3 · Choosing Your AI Surface & Workflow

Your First Cloud Sandbox

Last reviewed

Beginner

What you'll learn

~15 min
  • Set up the zero-install option you chose in the previous lesson
  • Complete a simple task end-to-end in your sandbox
  • Understand what persists and what doesn't in your environment
  • Know your two forward paths: continue through Module 4 and Module 6, or jump to Module 7
ℹAlready installed an agent in Lesson 1.2?

Already installed an agent on your computer in Lesson 1.2? You can skim or skip Lessons 3.4 and 3.5; they are here for locked-down work laptops, Chromebooks, and zero-install setups. If your sandbox gives you a terminal, Module 4 teaches the commands you need.

You chose your sandbox in the previous lesson. Let’s set it up and get your first task done.

Pick the path below that matches your choice. If you haven’t decided yet, go back to Choosing Your Cloud Sandbox. Path 0 is a no-repo browser builder; Paths A to C are repo-backed cloud agents and Linux devboxes.

Path 0: A no-repo browser builder (Claude Artifacts or ChatGPT Work & Sites)

No GitHub account and nothing to install: you work in a web page and share the result by link.

Setup

  1. Go to claude.ai or chatgpt.com and sign in.
  2. On chatgpt.com, switch the conversation from Chat to Work. On claude.ai, just start a new conversation; Claude opens an artifact beside the chat when you ask it to build something.

Your first task

Type:

Make a one-page sign-up sheet for a book club meeting: name, email, and which of three dates suits them. Keep it simple and friendly.

You should see the page appear beside or inside the conversation. Try it, then ask for one change (“Add a short welcome line at the top”), and watch it update.

What persists

The page lives with the provider, attached to your account and the conversation; share it with the share or publish button. Nothing is saved on your computer and there is no project folder, so this path suits quick tools rather than projects you will keep building. Check current plan limits in the cheat sheet.

Path A: Claude Code on the web

Claude Code on the web gives you Anthropic’s agent in your browser, connected to your GitHub repositories.

Setup

  1. Go to claude.ai/code (requires a Pro, Max, Ultra, or Team plan; on Enterprise, a premium or Chat + Claude Code seat)
  2. Sign in with your claude.ai account
  3. Connect your GitHub account when prompted
  4. Select a repository to work with — or create a new one on GitHub first

Your first task

Once connected to a repo, type:

Create a file called hello.md with a brief self-introduction. Include my name (use a placeholder), what I'm learning, and today's date.

Claude Code on the web will create the file, commit it, and push it to a new branch of your repo, not to your main branch. Check the Changes card beside the chat, or switch to that branch on GitHub to see the file. To bring it into your main branch, click Create PR and merge the pull request on GitHub.

What persists

Everything Claude pushes to GitHub persists. When you step away, the cloud machine pauses with its files saved, but after a longer period of inactivity it is reclaimed and only your conversation comes back. Anything not pushed is lost then, so check that your branch shows up on GitHub before you close the tab.

Path B: GitHub Codespaces

Codespaces gives you a full Linux terminal in your browser — the most flexible cloud option.

Setup

  1. Go to github.com/codespaces
  2. Under Explore quick start templates, find Blank (click See all if you don’t see it)
  3. Click Use this template
  4. Wait a moment while it starts: VS Code opens in your browser with a terminal
ℹDetailed Codespaces setup

If you need step-by-step instructions, see Lesson 4.5: Remote Environments; the Codespaces section walks through every step. This lesson assumes you have a running codespace.

Verify your tools

Type these commands in the terminal at the bottom to verify the underlying engines are ready. Don’t worry about what they mean right now — you’re just checking that everything is in place:

Terminal window
node --version # v22 or higher if you install tools with npm
npm --version # Should show a version number
git --version # Should show a version number

Install a CLI tool

Pick one tool to install. Claude Code is recommended for your first; Anthropic’s native installer needs no Node.js:

Terminal window
curl -fsSL https://claude.ai/install.sh | bash
claude --version

Or GitHub Copilot CLI (works with the free Copilot tier, no API key needed):

Terminal window
curl -fsSL https://gh.io/copilot-install | bash
copilot --version

Your first task

Create a project folder, move into it, and start the tool:

Terminal window
mkdir my-first-project
cd my-first-project
claude --dangerously-skip-permissions

The flag turns off the approval prompts, so the agent works without stopping to ask before each step. In a throwaway codespace that is a sensible default; Module 6 explains what the setting does on each tool. If you installed Copilot CLI instead, the equivalent start line is:

Terminal window
copilot --allow-all

Once the agent starts, switch to Plan mode (Shift+Tab or /plan), then type:

Create a file called hello.md with a brief self-introduction. Include my name (use a placeholder), what I'm learning, and today's date.

Read the short plan it proposes and tell it to go ahead.

Exit the tool (type /exit, or press Ctrl+C twice), then verify:

Terminal window
cat hello.md

What persists

Files in your codespace persist across sessions until the codespace is deleted (auto-deletes after extended inactivity; see the cheat sheet for current limits). Tools you installed (Claude Code’s and Copilot CLI’s native installs in your home folder) persist within the same codespace. Push important work to GitHub: codespaces are not permanent backups.

Path C: Google Cloud Shell

Cloud Shell provides a free browser-based terminal. Best for practicing terminal commands — limited for full development.

Setup

  1. Go to shell.cloud.google.com
  2. Sign in with your Google account
  3. A terminal appears — you’re ready
ℹCloud Shell details

See Lesson 4.5: Remote Environments for more on Cloud Shell’s capabilities and limitations.

Your first task

Terminal window
mkdir my-first-project
cd my-first-project
echo "# Hello from Cloud Shell" > hello.md
echo "I'm learning AI CLI tools. Today is $(date +%Y-%m-%d)." >> hello.md
cat hello.md

What persists

Files in your home directory (~/) persist across sessions. Install Node via nvm (which lives in ~/) so CLI tools persist too. Sessions time out after inactivity. See the Cloud Sandbox Cheat Sheet for current limits and full details.

Checkpoint: verify your sandbox works

Regardless of which path you chose, confirm you can do these three things:

  1. Create something: hello.md in a project folder (Paths A to C) or a working page (Path 0)
  2. Check it: read the file back, or try the page, and confirm it is what you expected
  3. Understand persistence: know whether your work will be there tomorrow

If all three check out, your sandbox is ready for real work.

What’s next: two valid paths

You now have a working zero-install environment. From here, you have two options:

Option 1: Continue in order: Module 4: Command Line Basics teaches the terminal your codespace or Cloud Shell gives you, and then Module 6, Your Agent In Depth: Desktop App & CLI, covers each tool (Claude Code, Antigravity, Codex, Copilot) in detail, as a desktop agent app and as a terminal CLI. Each lesson includes a “Cloud Equivalent” callout showing you the browser-based alternative. If you’re in Codespaces, you can install and use the CLIs directly. This is the recommended path, and the right one if you started with a no-repo builder and want to move on to real projects.

Option 2: Jump to Module 7, Git: If you’re using Claude Code on the web exclusively and don’t need to learn other tools right now, skip ahead to Module 7: Git. Git is the next essential skill regardless of which tool you use.

ℹDomain tracks need a workspace agent

The domain track lessons need an agent working in a project folder, for file operations, package installation, and project scaffolding. A desktop agent app or terminal CLI on your own computer works, and so do the CLIs installed in your codespace. A no-repo builder is not enough for these tracks.

🏛️In Your Field: Government Devclick to expand

Data sovereignty warning: All cloud sandboxes process your code on external servers. Do not paste sensitive, PII, CUI, or classified data into public sandboxes without authorization from your security office. If your agency’s network blocks certain cloud services, try each sandbox URL from your work machine before committing to a path. Codespaces and Cloud Shell are often accessible on networks that block direct software installation, since they run entirely in the browser. Your IT team may also have approved cloud development environments — ask before assuming you’re locked out. A sanctioned environment is not a CUI determination: CUI must not go into AI systems that are internet-enabled or external to agency control (ISOO Notice 2026-01). See approved tools and endpoints.

📊In Your Field: MIS / Businessclick to expand

For class projects, Codespaces is the strongest choice: verified students get free use up to 180 core hours a month (GitHub Classroom docs), the environment follows you across machines, and your professor can provide a template repository. If your university lab blocks GitHub, Cloud Shell is a fallback for terminal practice with a personal Google account (school-issued Google accounts often have it switched off), and ChatGPT data analysis handles single-file data tasks.

🧬In Your Field: Biotechclick to expand

If your lab uses shared workstations, Codespaces gives you a personal, consistent environment for bioinformatics scripting. Install Python, BioPython, and any CLI tools you need; they persist within your codespace. For quick data processing (parsing CSVs, reformatting FASTA files), ChatGPT data analysis works without any setup: upload the file, describe the transformation, download the result.

KNOWLEDGE CHECK

You're on a Chromebook and want to build a multi-file website project that you will keep in a GitHub repository. Which option is the best fit?

KNOWLEDGE CHECK

What is the main tradeoff of zero-install options compared to an agent on your own computer (desktop app or CLI)?

Key Takeaways

  • Your zero-install option is set up and working: you’ve created and checked your first result
  • No-repo builders (Claude Artifacts, ChatGPT Work & Sites) make a shareable page fast, but there is no project folder to keep building in
  • Claude Code on the web connects to GitHub repos: its work lands on a branch you can turn into a pull request, and anything not pushed can be lost once the session goes idle
  • Codespaces is a full Linux machine — install CLI tools, run commands, persist files across sessions
  • Cloud Shell is a free terminal for practice: files and tools installed in ~/ (such as Node via nvm) persist, but anything installed outside ~/, for example with sudo apt install, is gone after the machine shuts down
  • Two valid forward paths: continue in order through Module 4 and Module 6, or jump to Module 7 if using Claude Code on the web exclusively
Search lessons