This session stands alone. A camper who attends only this session leaves with a working app they built, understands the loop that produced it, and could do it again. Nothing here assumes a Session 2.
For middle school this is the entire track — one 85-minute session, and it ends here. High school continues into Sessions 2–4.
MS and HS run the same seven blocks. HS gets 35 more minutes, spent almost entirely on independent build time and on the concepts blocks.
| Time | Block | Mode |
|---|---|---|
| 0:00–0:08 | Hook: build an app in front of them | Instructor demo |
| 0:08–0:15 | What just happened | Short talk |
| 0:15–0:40 | Build #1: everyone builds Mad Libs | Guided, lockstep |
| 0:40–0:48 | The three moves: Ask, Run, Fix | Short talk + demo |
| 0:48–0:53 | Break | |
| 0:53–1:10 | Build #2: make it yours | Independent |
| 1:10–1:18 | The AI was wrong + break your app | Whole group |
| 1:18–1:25 | Showcase + take it home | Campers demo |
Take the break. This ran as written with a real MS group — full 85 minutes, break included, workload about right. If you’d rather skip it, give the 5 minutes back to Build #2 (making it 22) — but don’t spend them on the 1:10 or 1:18 blocks.
| Time | Block | Mode |
|---|---|---|
| 0:00–0:10 | Hook: build an app in front of them | Instructor demo |
| 0:10–0:22 | What just happened | Short talk |
| 0:22–0:50 | Build #1: everyone builds Mad Libs | Guided, lockstep |
| 0:50–1:05 | The three moves: Ask, Run, Fix | Short talk + demo |
| 1:05–1:40 | Build #2: pick your own app | Independent |
| 1:40–1:52 | The AI was wrong + break your app | Whole group |
| 1:52–2:00 | Showcase + take it home | Campers demo |
Do not explain anything first. Open a terminal on the projector and build something.
cd $HOME\Documents\Projects
mkdir <Name>
cd <Name>
mkdir demo
cd demo
claude
Make your own name folder for <Name> — the same one campers will use. First
name, one word, matching their Google Drive folder.
Type your prompt out loud so they see you thinking:
Build a dice roller in Python with a tkinter window. Big button that says ROLL, and when I click it, it shows a random number from 1 to 20 in huge text. Save it as
dice.py.
Let it generate. Narrate what you see scroll by — “it’s writing a file, it’s showing me the code.” Then run it:
python dice.py
Click the button a few times. Then, live:
Make the number change color — green if it’s 20, red if it’s 1, black otherwise.
Run it again. Roll until you get a 20.
Total elapsed: under 6 minutes. Spend the remaining time letting them react, then:
“That’s the whole class. You describe, it builds, you run it, you fix it. Everything else today is getting good at those four things.”
MS/HS
MS: Take one request from the room for a third change. The visual payoff lands hard and it buys you their attention for the next 25 minutes. HS: After the demo, show them the actual
dice.pyfile. Ask: “Anything in here surprise you? Anything you’d have done differently?” Sets up Session 2 verification.
Keep this tight. Three ideas, no more.
The AI writes the code. You decide what gets built, whether it’s good, and what changes. That’s a real job — it’s most of what senior engineers actually do.
It has read an enormous amount of code and it’s very good at predicting what code should come next. That’s it. It is not thinking about your project. It doesn’t know what you meant unless you said it. It will confidently produce something wrong, and it will look just as polished as something right.
“It’s like a genius intern on day one. Incredibly fast, knows every library, has no idea what you actually want, and will never tell you it’s confused.”
Code that looks right and doesn’t run is worth nothing. You’ll run your app more times today than you’d guess.
Point at the terminal:
claude starts Claude Code, an AI that can read and write files in the folder
you’re in and run commandsCtrl+C gets you out. claude gets you back in.Say
claudefor the rest of the session. One command — don’t introduce a second.
MS/HS
MS: Stop after the intern analogy and the tool basics. Skip the tool/model distinction entirely — it costs 3 minutes you don’t have. HS: Worth 2 minutes on tool-vs-model. It’s a genuinely useful mental model for how AI products are built, and it explains why different setups behave differently.
Lockstep. Everyone builds the same thing so nobody is lost and everyone succeeds early.
cd $HOME\Documents\Projects
mkdir <Name>
cd <Name>
mkdir madlibs
cd madlibs
claude
<Name> is the camper’s name folder — first name, one word, matching their
Google Drive folder. Walk the room. This is where setup problems surface — catch
them now.
Put this on the projector. Campers type it themselves (typing it makes them read it):
Build a Mad Libs generator in Python with a tkinter window.
Ask the user for 5 words: a noun, a verb, an adjective, a place, and a food.
Each one gets its own text box with a label.
When they click a "Make Story" button, show a funny story using their words.
Save it as story.py
Then:
python story.py
Everyone stops here until everyone’s app runs. Fast finishers help neighbors — this is a real job too.
Put the prompt back up and mark it up:
story.py“Four things: what kind of thing, what it does, what it looks like, where to put it. That’s a good prompt. Notice what’s not there — nothing about how to write the code. That’s the AI’s job.”
Everyone makes one change of their own choosing. Suggestions on the board:
Rule: make the change, run it, then stop. Hands up when yours works.
This is the conceptual core of the session. Everything else is practice.
A good ask answers four questions:
| Question | Weak | Strong |
|---|---|---|
| What kind of thing? | “make a timer” | “a Python tkinter app” |
| What does it do? | “counts down” | “counts down from 25 minutes, beeps at zero” |
| What does it look like? | (nothing) | “big digits in the middle, Start and Reset buttons” |
| Where does it go? | (nothing) | “save it as timer.py” |
The size rule. The single most important thing today:
Ask for one thing at a time.
Not because the AI can’t do five things — it often can. Because when five things come back at once and one is broken, you don’t know which one. When you ask for one thing and it breaks, you know exactly where to look.
When you should ask for a lot at once: the very first prompt. Starting from nothing, describe the whole app. After that, one change at a time.
python story.py
Run after every single change. Not “every few changes.” Every one.
Three things can happen:
Number 3 is where humans matter. The computer can’t tell you your Mad Libs story is boring, that your timer counts up, or that your quiz says you got 3/5 when you got 4/5. Only you can.
How to report a problem, worst to best:
Say what you did, what you expected, and what actually happened.
That’s not an AI trick — that’s how you report a bug to any human on earth, forever. It’s a real professional skill and they’re learning it today.
MS/HS
MS: Cover the size rule and Move 3 properly; compress Move 2’s three outcomes to “works / crashes / weird.” Skip the four-question table as a table — just say the four things out loud while pointing at the Mad Libs prompt still on screen. HS: Full version, plus the demo: ask for five changes at once, let it come back tangled, then show the same five done one at a time. The difference is obvious and they’ll remember it. Then push on outcome 3 — “How would you know your app is wrong?”
This is where the two age groups genuinely diverge in ambition.
MS: make it yours. Campers keep working on story.py and customize it heavily —
their own story templates, inside jokes, colors, extra buttons, a random-story mode.
Anyone who finishes early and wants a fresh app may start one from
project-ideas.md, Starter tier only.
HS: pick your own app. Campers choose from project-ideas.md
or bring their own. Instructor approves the idea before they start — sanity check on
scope. Starter or Solid tier.
Everyone writes their first prompt (or their next three changes, for MS) on paper and gets it checked against the four questions. This catches the “make me a game” prompts before they waste the whole block.
Independent work. Instructor circulates.
MS is tight here — 10 minutes is enough to make three real changes to story.py and
no more. Anyone starting a fresh app from the idea bank needs to have decided that during
the previous block, not now.
Checkpoints on the board:
Instructor moves while circulating:
Ctrl+C, delete the file, start fresh with a better
first prompt. Restarting is cheap and this is a valuable lesson.Everyone stops building. Last change gets tested, and they write one sentence: “My app lets you ___.”
MS/HS
MS: Pairs. Call a checkpoint at the halfway mark — “hands up if your app runs right now.” Anyone with a hand down gets you next. HS: Solo. Add a checkbox: “I read the code and understood one part of it.”
The most important block of the session. Do not cut this for time.
Put this on the projector — a real thing an AI will produce for “make a quiz app that shows my score at the end”:
score = 0
for question in questions:
answer = input(question["text"] + " ")
if answer == question["correct"]:
score += 1
print(f"You got {score} out of {len(questions)}!")
print(f"That's {score / len(questions) * 100}%!")
Ask the room: “What’s wrong with this?”
Let them find it. Seed hints only if they stall:
33.33333333333333%)None of this crashes. It runs perfectly. It’s just wrong, and it will hand a camper a score they didn’t earn.
Then land the point:
“The AI wrote code that runs. It didn’t write code that’s right. Nothing in your computer will ever tell you the difference. That’s your job — and it’s why this class exists.”
Then, hands-on: everyone spends the last few minutes trying to break their own app.
Found something? Fix it. That’s the last change of the day.
MS/HS
MS: Do the breaking exercise first (it’s fun and physical), then show the code example with only the case-sensitivity bug. Skip the percentage and whitespace bugs. HS: Ask them to write the fix for each bug before you show it. Then ask the better question: “How would you have caught this if I hadn’t told you?” Answer: you test with the wrong input on purpose.
There is not time for everyone to present. Run a gallery walk: everyone leaves their app running, campers circulate for 3 minutes, then 3–4 volunteers show theirs to the whole room for 45 seconds each.
Volunteers cover two things:
Item 2 is non-optional. Making the struggle public normalizes the loop and defuses the idea that anyone got it right the first time.
Three things they leave with:
Ask small → Run it → Say exactly what went wrong → Repeat
Their app. It’s theirs. Have them copy it to their Google Drive folder — the link is in the camper notes.
“You built something real today. Also — the AI wrote the code, and you should know what that does and doesn’t mean. It means you can build things right now that would have taken you months to learn. It doesn’t mean you know how the code works. If you want to actually get good at this, the next thing to do is read the code the AI wrote and figure out what each part does.”
MS/HS
MS: Trim the honest note to its first two sentences. Then: “You made a thing that works. Show somebody.” This is the last thing middle school hears all camp — land it, and make sure every app is copied to Drive before anyone leaves. HS: Full version, plus: “The people who can do both — steer the AI and read the code — are going to be unbelievably good at this.”
HS only — they’re back for Session 2, so give them the teaser: “Next time we find out how to tell when the AI is lying to you.” Middle school finishes here; don’t tease a session they won’t attend.
ms-camper-notes.md or hs-camper-notes.md, one per
camper. Hand out at the start; they’re written to be followed along with.claude on a lab machine the morning ofDocuments\Projects folder exists on every lab machine — create it
if it’s missing<Name> is the camper’s first name, one
word, matching their Google Drive folder. Expect questions — it’s new.python -c "import tkinter" (no output = good)claude before campers arrivedice.py demo working before campers arriveCtrl+C out of a hung session and restart — you’ll do it a dozen timestroubleshooting.md
This Vibe Coding Curriculum by Brian Bird, created in with AI assistance, is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.