AI HORIZON
For Instructors

Set Up Your Own AI Coding Assistant

How to build and edit notebooks (like the ones in this workshop) by describing what you want in plain English — using VS Code + Claude Code, the same setup used to create these materials.

⏱ ~30 minutes, one time 💻 Windows or Mac 🤖 No coding required to start

What this gives you

An AI assistant that lives inside your code editor. The real unlock for an educator: you chat about a concept you want to teach, and it builds the code and the step-by-step notebook to walk your students through it. You describe what you want — "make a notebook that compares three tools," "add a progress bar," "explain this error" — and it writes and edits the files for you, showing you each change to approve. It's how every notebook and handout in this workshop was built.

💳

Read this first: it costs money

The AI assistant needs either a paid subscription or pay-as-you-go credits. For Claude Code that's a Claude Pro ($20/mo) or Max ($100–$200/mo) plan (or Anthropic API credits). Already pay for ChatGPT? You can use OpenAI Codex on your existing ChatGPT Plus/Pro/Team plan instead — see the OpenAI section below. The VS Code editor and the extensions are free; only the AI usage is paid. The $20/mo tier of either is plenty to start.

What you'll install

  • Python — the language the notebooks run
  • VS Code — the code editor
  • Two VS Code extensions — Python & Jupyter
  • Node.js — required by the Claude Code tool
  • Claude Code — the AI assistant (CLI + VS Code extension)
1

Install Python

Download from python.org/downloads and run the installer.

Windows — important

On the first installer screen, check the box "Add Python to PATH" before clicking Install. (Skipping this causes most "python not found" problems.)

Verify (any OS)

Open a terminal — Windows: PowerShell · Mac: Terminal — and run:

python --version

You should see a version like Python 3.12.x. Mac tip: if python isn't found, try python3 --version.

2

Install VS Code

Download from code.visualstudio.com and install it (accept the defaults). VS Code is the editor you'll work in.

3

Add the Python & Jupyter extensions

Open VS Code. Click the Extensions icon in the left bar (or press Ctrl/Cmd + Shift + X). Search for and install:

  • Python — publisher Microsoft
  • Jupyter — publisher Microsoft

These let you open and run .ipynb notebooks right inside VS Code.

4

Install Node.js

Claude Code runs on Node.js. Download the LTS version from nodejs.org and install it (accept defaults). Then verify:

node --version

You should see something like v20.x or higher.

5

Install the Claude Code tool

In your terminal, install Claude Code globally:

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

Confirm it installed:

claude --version
6

Add the VS Code extension & sign in

Back in VS Code's Extensions panel, search "Claude Code" and install the one by publisher Anthropic. Note: needs VS Code version 1.98 or newer (use Help → Check for Updates if unsure).

Open the Claude Code panel and start it. The first launch opens your web browser to sign in to your Claude account (the OAuth flow). Approve it, and you're connected.

🔑

Two ways to sign in

Subscription (recommended to start): sign in with your Claude Pro/Max account in the browser when prompted. API key: alternatively, set an ANTHROPIC_API_KEY from console.anthropic.com for pay-as-you-go billing.

Already have a paid ChatGPT account? Use OpenAI Codex instead

If you're already paying for ChatGPT, you don't need a Claude subscription — OpenAI's coding agent, Codex, works the same way inside VS Code. Steps 1–4 above are identical (Python, VS Code, the Python & Jupyter extensions, Node.js). Then, instead of Claude Code:

A

Install the Codex extension

In VS Code's Extensions panel, search "Codex" and install the official one by publisher OpenAI (Extension ID openai.chatgpt).

Optional command-line tool

npm install -g @openai/codex
B

Sign in with ChatGPT

Click the new Codex icon in the activity bar to open its panel, then click "Sign in with ChatGPT." Your browser opens, you log in to your ChatGPT account (the OAuth flow), and Codex connects automatically.

🔑

What you need

A paid ChatGPT plan (Plus $20/mo, Team, Pro, or Enterprise) — Codex usage is included in your plan, no extra setup. Or sign in with an OpenAI API key for pay-as-you-go billing.

🟰

Same workflow either way

Claude Code and Codex work almost identically — chat in plain English, review the proposed diffs, accept or reject. Everything below applies to whichever one you chose. Pick one and stick with it while you learn.

Now use it — the actual workflow

This is exactly how the workshop notebooks were made (Claude Code or Codex — same idea):

Your first task

1. In VS Code: File → Open Folder, and pick a folder to work in (even an empty new one).

2. Open the Claude Code panel and type a request in plain English, for example:

Create a Jupyter notebook that loads a small dataset with pandas and makes a simple bar chart. Explain each step in markdown cells.

3. Claude proposes the files and changes as inline diffs — you review and click Accept or Reject. Nothing changes without your OK.

4. Keep the conversation going: "add a second chart," "explain that error," "make it use a different dataset." It remembers the context.

💡

Great starter exercises

Clone this workshop, open it, and ask Claude to "explain what notebook 04 does" or "add a new prompt to the benchmark." Learning by modifying real, working material is the fastest way in.

Building notebooks that run on Jetstream2

A notebook that works on your laptop can break on a fresh Jetstream2 instance — different OS, nothing pre-installed, and usually no GPU. When you ask the assistant to build a teaching notebook for the cloud, give it that context so the result is cloud-ready. Paste something like this into your request:

This notebook will run on a FRESH Jetstream2 Ubuntu instance (often CPU-only, on a NAIRR allocation). Make it self-contained and cloud-ready: - put all pip/system installs in the first cells (the instance starts bare) - auto-detect CPU vs GPU and adapt (do NOT assume a GPU is present) - use a SMALL model so it fits the default disk and the allocation's SUs - explain each step in markdown cells for students - add one setting at the top for a quick "demo" run vs a full run
  • Self-contained installs. The instance is blank — the notebook must install what it needs in its opening cells.
  • CPU-aware. Most class instances are CPU-only; have it detect the hardware and skip GPU-only steps automatically.
  • Small footprint. Large models overflow the disk and burn SUs — ask for small models and a quick "demo mode."
  • Push to GitHub. Ask the assistant to commit and push; students then git clone it on their instance (that's Track A of the workshop).
🔁

The full teaching loop

You describe a lesson → the assistant builds the notebook → you push it to GitHub → students clone and run it on Jetstream2. Notebook 04 in this very workshop was built exactly this way.

📝

Beyond notebooks: proposals & reports too

The same assistant helps with writing, not just code — drafting your full NAIRR proposal, a final report, a syllabus, or a workshop handout. Describe what you need and it drafts sections you refine. (It's exactly how this workshop's guides and proposal templates were written.)

Other options (if you're curious)

We recommend starting with Claude Code because it's what these materials were built with — but the landscape has good alternatives. Pick one and stick with it at first; switching tools mid-learning is the main way people get stuck.

ToolNotes
Claude Code (recommended)Anthropic's assistant. VS Code extension + terminal. Subscription or API credits.
GitHub CopilotDeeply built into VS Code; uses OpenAI/Anthropic models. Subscription ($10/mo), free tier available.
OpenAI Codex / ChatGPTOpenAI's coding tools, including a CLI and IDE options. Subscription or API credits.
OpenCodeOpen-source, terminal-based AI coding agent. Free software; you still bring your own model/API key.
🔒

A word on responsible use

These tools send your prompts (and the files you reference) to the AI provider. Don't paste secrets, student data, or anything confidential. For teaching materials and open code, you're fine.

NAIRR Workshop Series · Workshop 01 — Instructor Setup Guide · Build your own with AI
Part of the AI Horizon project · NSF #2528858 · CSUSB Center for Cyber and AI