AI Engineering Academy Monmouth University · HH216 · Summer 2026
PHASE 01 · OPEN

AI Engineering Academy

Build a local web page with an AI teammate. No coding background required. 125 min.

the orbit is your agenda — each phase rotates one notch

← → navigate · I ink-wash · F fullscreen

PHASE 01 · OPEN · 0:00 — 16:00
PHASE 01 · OPEN

5-minute quiz

No grade. We measure what you learn today.

Open the web course on your laptop:

weihaoqu.github.io/learnai-course

Pre-test is the first step inside.

PHASE 01 · OPEN · 0:00 — 16:00
PHASE 01 · OPEN

Read. Decide.

I Agree or I Do Not Agree — either way you do everything today.

Inside the web course:

weihaoqu.github.io/learnai-course

Read the consent. Click your choice.

Non-consent path → facilitator runbook (NOT shown to students)

PHASE 01 · OPEN · 0:00 — 16:00

— PHASE 01 · OPEN

Setup in four checks.

Copy only the text after the $. If any check fails, raise your hand.

01 · Open

Open Terminal. This is where commands go.

02 · Node

node --version
If not found: nodejs.org → LTS.

03 · Install

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

04 · Start

claude
Log in when it asks.

~/workshop
$ node --version
v22.11.0
$ npm install -g @anthropic-ai/claude-code
✓ claude-code installed successfully
$ claude

Do not type the $. It only means Terminal is ready.

Facilitator: Demo slowly. Confirm Terminal, Node, install, and first Claude prompt. If someone is stuck for more than 90 seconds, pair or move them to the browser fallback.

PHASE 01 · OPEN · 0:00 — 16:00

— PHASE 01 · OPEN

You're ready when Claude answers.

Claude Code
> claude
Hi! I'm Claude. I can read your files, write code, run commands,
and build things with you. What are we making today?
Success

You see a > prompt and Claude asks what to build.

Stuck

Raise your hand. Do not keep retrying random commands.

Fallback

Use claude.ai: ask for one self-contained HTML file, save it, open it locally.

This is your agent. It edits files, runs code, and builds things under your supervision.

PHASE 01 · OPEN · 0:00 — 16:00
PHASE 02 · SAFETY

Read this before you talk to Claude.

❌ NO real name (use a nickname)
❌ NO address, school name, phone, email
❌ NO passwords, API keys, health, family details
❌ NO face photos or screenshots with personal info
❌ NO surprise installs after setup: npm, brew, pip, git clone
✅ OK: create, edit, and open today's local HTML file
✅ Pause on prompts that mention delete, network, login, or secrets

Parents: no screen photos · no coaching answers · no pressure around consent.

PHASE 02 · SAFETY · 16:00 — 21:00
PHASE 02 · SAFETY

Pick your nickname.

Write it on a sticky note. We use it for the whole session.

Skyline · NebulaCat · Trail · GameDev42 · null

PHASE 02 · SAFETY · 16:00 — 21:00
PHASE 03 · QUICK WIN

First page. Right now.

Paste this. Hit enter. Watch.

Create one local file named starter.html. Use plain HTML and CSS only. No React, Vite, Next, npm installs, or API keys. The page should say: "Hi, I'm <nickname>". Dark background. One sentence about what my persona likes (anything fictional is fine). Open it in my browser when done.
0 / 5 pages rendered
10:00

Goal: every student has a page on screen by minute 25.

PHASE 03 · QUICK WIN · 21:00 — 29:00

— PHASE 04 · DECODE

claude.ai

Chat. Answers. Explains.

Edit your files
Run terminal commands
Remember your project
Install skills
Build across a codebase
Answer questions

Claude Code

Agent. Acts. Builds.

Edit your files
Run terminal commands
Remember your project
Install skills
Build across a codebase
Answer questions
PHASE 04 · DECODE

Claude reads CLAUDE.md first.

── CLAUDE.md ────────────────── # Workshop Notes - Call me <nickname> - Beginner mode: explain before changing code - Plain HTML/CSS only; no installs - Ask before deleting, networking, or using secrets ──────────────────────────────

A project note for your AI teammate. Guidance, not enforcement.

PHASE 04 · DECODE · 29:00 — 39:00
PHASE 04 · DECODE

Permission prompt = pause.

  1. Usually OK — create, edit, or open your local HTML file.
  2. Ask Q first — install, delete, network, login, API key, password.
  3. Reset/clear starts a fresh conversation, not a fresh file.

Read the file path and command out loud before approving.

PHASE 04 · DECODE · 29:00 — 39:00
PHASE 05 · PLAN

Plan beats Prompt.

The #1 mistake: jumping in too fast.

PHASE 05 · PLAN · 39:00 — 57:00
PHASE 05 · PLAN

PLAN → ONE PROMPT → ITERATE → POLISH

You'll do all four — same way pros do.

PHASE 05 · PLAN · 39:00 — 57:00
PHASE 05 · PLAN

Make Claude interview you.

Today: no-install prompt. Later: Q's reusable interview skill.

Interview me one question at a time to plan a personal landing page for my nickname persona. Cover: who they are, sections, vibe, color, signature, edge cases, and done checks. Output a structured plan.

The reusable version is project-spec-interviewer: vague idea -> implementation-ready spec.

→ Next: watch it happen.

PHASE 05 · PLAN · 39:00 — 57:00

— PHASE 05 · PLAN

Watch it happen.

Claude Code — project-spec-interviewer pattern
Let's turn this into an implementation-ready plan. First question:
What do you want someone to be able to do with what you build?
> I want visitors to understand my fictional space-designer persona
··· thinking
Nice. Next question:
What sections should the landing page have?

One question at a time. Build only after the done checks are visible.

Facilitator cue:
Student volunteer on your laptop.
Prompt: "Use the project-spec-interviewer pattern. Ask one question at a time."
Let it surface 2–3 done checks. Then move on.
PHASE 05 · PLAN

A good plan has:

  • 1 sentence about your persona
  • 3 sections (e.g., About / Projects / Contact)
  • A vibe word (calm / bold / playful / pro)
  • A color
  • One thing that makes it YOURS
PHASE 05 · PLAN · 39:00 — 57:00
PHASE 05 · PLAN

Show Q your plan.

30-second review. Wave me over.

0 / 5 plans approved

No good plan, no build.

PHASE 05 · PLAN · 39:00 — 57:00
PHASE 06 · BUILD

Now the real one.

Using the plan we just made, create one local file named landing.html. Use self-contained HTML, CSS, and vanilla JavaScript only. Do not install packages. Do not use React, Vite, Next, API keys, or accounts. Open landing.html in my browser when done.

Why one prompt

Small site = one-prompt scaffold for learning. Real projects = small prompts, verify each step.

PHASE 06 · BUILD · 57:00 — 75:00
PHASE 06 · BUILD

Test before you trust.

  • Does the file open in the browser?
  • Does it match your plan?
  • Do links, buttons, and text look right?

One bug found = one better prompt.

PHASE 06 · BUILD · 57:00 — 75:00
PHASE 06 · BUILD

Talk like a teammate.

"The page opens, but the header is too small. Make only the header bigger." "The colors feel harsh. Try a calmer palette. Keep it one HTML file." "Add a dark mode toggle. Do not install anything."

Plain English. Specific. One at a time.

PHASE 06 · BUILD · 57:00 — 75:00
PHASE 07 · POLISH

Use open components.

21st.dev · shadcn/ui — free, open licenses.

Check the license

Don't paste closed-source code. Open licenses are clearly labeled (MIT, Apache, etc.).

PHASE 07 · POLISH · 75:00 — 87:00
PHASE 07 · POLISH

5 cheats to look pro.

  • Gradient background
  • One sans-serif font (Inter, Geist)
  • Lucide icons, not emojis
  • Generous whitespace
  • Dark mode by default

Tell Claude: "apply these."

PHASE 07 · POLISH · 75:00 — 87:00
PHASE 07 · POLISH

Second opinion.

Different AI. Different blind spots.

/codex review

Important

Review reduces risk. It does not prove correctness.

PHASE 07 · POLISH · 75:00 — 87:00
PHASE 07 · POLISH

Give Codex a persistent direction.

Interview first; /goal second. Objective + completion conditions.

In Codex, type: /goal <objective> so it is done when <checks pass> Case study: these slides /goal improve these slides so HS beginners can write a /goal, faculty can reuse the workflow, research students can track evidence, and checks pass: clean layout, self-contained, no JS errors. Your turn: /goal build my science fair page so it opens locally, explains my idea, and has one checklist I can test.
PAUSE

Pause before starting when audience, constraints, or done checks are missing.

MOVE ON

Proceed when the output is visible and the checks are clear and testable.

FINISH

Stop or rewrite the goal when checks pass or the direction has changed.

/goal is experimental direction, not autopilot. You still review every step.

PHASE 07 · POLISH · 75:00 — 87:00
PHASE 08 · SHARE

2 min each.

  • Show your page
  • One thing you learned
  • One thing AI got wrong
Best AI fumble of the day

Verbal award only — no name on screen. Criteria: funniest bug narrative, not most embarrassed student.

PHASE 08 · SHARE · 87:00 — 93:00
PHASE 09 · LEVEL UP

Pre-packaged superpowers.

── .claude/skills/landing-page/SKILL.md ──
---
name: landing-page
description: Build a personal landing page
---
Interview me one question at a time.
Cover: persona, sections, vibe, color.
Then build the HTML file.
──────────────────────────────────────

Did it twice? Codify it. Type /landing-page anywhere.

Project: .claude/skills/…/SKILL.md · Personal: ~/.claude/skills/…/SKILL.md

→ Next: install one right now.

PHASE 09 · LEVEL UP · 93:00 — 115:00

— PHASE 09 · LEVEL UP

Verified setup command · facilitator only

~/setup
$ npm install -g @weihaoqu/project-spec-interviewer-skill

Students: do not type this live unless Q or the facilitator says it is part of setup.

What it gives you

Claude Code
> Use project-spec-interviewer for my app idea.
I will ask one question at a time, then write the plan, edge cases, and acceptance criteria.

Interview skill = better specs. Better specs = cleaner plans, better /goal checks, fewer rebuilds.

PHASE 09 · LEVEL UP

MCP = connect to anything.

Protocol for external tools, data, and APIs.

[ Claude Code ] ─── MCP ─── [ Gmail · GitHub · Drive · custom ]

Today

We did NOT connect personal accounts.

PHASE 09 · LEVEL UP · 93:00 — 115:00
PHASE 09 · LEVEL UP

Set once. It runs without you.

  • Hooks — on every save
  • Cron — daily/weekly
  • Agents — background jobs

Pros set it once. The work happens without them.

PHASE 09 · LEVEL UP · 93:00 — 115:00
PHASE 09 · LEVEL UP

Research workspace.

After today: same pattern, higher stakes.

Rules

CLAUDE.md
AGENTS.md

Ask. Challenge. Name tradeoffs.

Evidence

notes/ sources/
drafts/ reviews/

Verify claims, paths, and counts.

Review loop

draft -> review -> fix -> verify -> record

Done means evidence, not confidence.

Working agreement

Builder agent drafts. Reviewer agent checks. You decide when the evidence is strong enough.

In our setup: Claude writes and revises; Codex reviews.

PHASE 09 · LEVEL UP · 93:00 — 115:00
PHASE 09 · LEVEL UP

Keep going in order.

Open the hub page, then read these pages top to bottom.

  • 1. Agentic Engineeringwhat-is-agentic-engineering
  • 2. Vibe Codingvibe-coding-beginners-guide
  • 3. Claude Code 101claude-code-101
  • 4. Agent Vocabularyai-agent-primer-ladder-workflow
  • 5. Roadmapagentic-ai-engineer-roadmap-2026

Then choose a route: teaching, agents, power-user workflow, or academic research.

weihaoqu.github.io/learnAIDoc/wiki/start-here-ai-agents-beginners/

PHASE 09 · LEVEL UP · 93:00 — 115:00
PHASE 10 · CLOSE

Same quiz. New score.

10 min. Honest answers shape the next session.

Same place — the web course:

weihaoqu.github.io/learnai-course

Post-test + 5-min survey at the end.

PHASE 10 · CLOSE · 115:00 — 125:00
PHASE 10 · CLOSE

Three things.

  • AI is fast. You are slow on purpose.
  • A plan beats a clever prompt.
  • You direct the toolchain.
PHASE 10 · CLOSE · 115:00 — 125:00
PHASE 10 · CLOSE

Thank you.

Build something dumb this week. Email it to me.

weihaoqu.github.io/learnai-course · weihaoqu25@gmail.com

PHASE 10 · CLOSE · 115:00 — 125:00
0/0