The map for the whole week
How founders get from problem to product
Problem
the thing that's broken
10 solutions
storm it with AI
Pick one
the smallest wedge you can win
One feature
the 90/10 cut
Screens
walk the user's journey
Build
aim the agent, ship it
You will run this exact path in the studio at 15:15.
Call-on · solution storm, live
NO LAPTOPS · I pick namesHungry students, 25-minute lunch, one crowded counter. Give me ten ways to fix it.
I cold-call. An app is one answer, not ten. Boring counts, weird counts, no-software counts. Then two more names: which ONE would a beginner team ship first, and what single feature proves it works?
Click: the exact prompt your team runs on YOUR problem at 15:15
"Give me 10 different ways a [your person] could solve [your problem], from boring to wild. For each, one line on what the simplest version looks like. Then tell me which one a beginner team could build a first version of in four days, and what its ONE first feature should be."
The cut, in ten seconds
Nine features die. One survives. That's the cut your team runs in the studio next.
StudyBuddy Ultra, unfiltered
Flashcards
Study forums
Shared calendar
AI tutor chat
Group video calls
Leaderboard + streaks
Dark mode / themes
Avatar shop (buy hats)
Parent progress dashboard
Quiz generator: paste notes, get 5 questions back
The rule
One feature. Cut everything that doesn't serve the core action this week.
Y Combinator calls this the 90/10 solution: get 90% of the value from 10% of the effort, and ship THAT.
A vague spec makes the agent build the wrong thing, fast, and put it on the internet before you notice.
What an agent actually is
It reads your files. Writes real code. Runs it. Deploys it. All of it, fast.
That speed cuts both ways. Hand it a vague instruction and it will happily build the wrong thing, fast, and put THAT on the internet too. The exact prompt you're about to give it is the only thing standing between you and a fast, wrong app. In the studio next, that instruction gets a name: spec v0.
Quick vocabulary
A terminal is just texting your computer without the pretty buttons.
Programmers love it because it is fast. You will see me use it for ten seconds to look cool. You will use Kimi Code in a terminal window; the setup page makes it a copy-paste job. Nobody needs the scary black screen this week.
First-contact demo · live, no rehearsal
LunchRush, live, on the internet, right now.
LIVE DEMO · no rehearsal
"Build a single-page website called LunchRush. One job: show which cafeteria line is shortest right now. A big headline with the current shortest line, a list of three stations with estimated wait minutes, and a refresh button. Plain and working beats pretty."
Real prompt, real deploy, live. If it breaks, it breaks live. Nothing about this is magic.
Just watch
Watch the terminal, not your own laptop.
Notice it reads files before it writes any.
Notice it explains its plan before acting.
Watch for the moment a real URL appears.
Your turn is next: your product, laptops open then.
Notice it reads files before it writes any.
Notice it explains its plan before acting.
Watch for the moment a real URL appears.
Your turn is next: your product, laptops open then.
What just went live
LunchRush: one page, one job
Same URL, same name, keeps showing up: cc-102 tomorrow, Day 3 again.
Mission 0 FOLLOW ALONG · laptops open
One laptop per team. Install Kimi Code.
5:00
1Open portal.arenaschool.org/kimi-setup on the team laptop. Every command is there with a Copy button.
2Install, open a fresh terminal, run kimi, then /login and paste the API key pinned in the class group chat. The key goes into Kimi Code and nowhere else, ever.
3Every teammate can see the screen. You build YOUR product starting now, not a demo. Hands up when you see the Kimi welcome screen. Go.
5:00
Every team in. Straight to the starter prompt.
Copy this prompt · fill in your problem
The starter prompt
Fill in your brackets
We are building [product name or working name] for [your person] who [your problem]. Build the first version of the main screen: what the user sees when they arrive, and the one thing they can do. Keep it simple and clean.
Filled in · LunchRush
We are building LunchRush for hungry students who have a 25-minute lunch and one crowded counter. Build the first version of the main screen: what the user sees when they arrive, and the one thing they can do. Keep it simple and clean.
Swap in your team's product, person, and problem. That's it. Run it.
This is what you're about to build
Your product's first screen, in thirty minutes
Not a team name. Not a thesis. The actual screen a real user would land on.
Sample prompts · pick and mix
Not sure what to type next? Steal one of these.
Inspiration
"Show me 3 different takes on this screen, then I'll pick one."
The iterate move
"I expected [X], it does [Y]. Fix that."
Design schema
"Apply a liquid glass design: frosted panels, soft shadows, rounded corners."
"Make it look like airbnb.com: same feel, my content."
"Dark, minimal, big type, like a modern startup site."
"Make it look like airbnb.com: same feel, my content."
"Dark, minimal, big type, like a modern startup site."
Safety
"Save a version before you experiment."
A checkpoint you can roll back to if the next ask makes it worse, not slower.
Exercise 1 · do it nowFOLLOW ALONG · laptops open
Build your first screen
30:00 on the clock
YOUR TOOLS· Build: Kimi Code (every laptop, no VPN) · v0 (school device backup) · Same starter prompt, same rules
You do, step by step
1Open your coding agent with your team's project.
2Give it the starter prompt, filled in with YOUR product. Deploy it.
3First screen renders? Save a version before you experiment.
4Post the URL, or a screenshot of the preview, in the class channel.
If you get stuck
Stuck 3+ minutes? Raise a hand, keep trying.
Still nothing at 20:00? Switch to Lovable, same starter prompt, no shame. Rendering is rendering.
Ask: what should this screen show, and what's the one thing someone can do on it? That gets you unstuck faster than the fix itself.
Still nothing at 20:00? Switch to Lovable, same starter prompt, no shame. Rendering is rendering.
Ask: what should this screen show, and what's the one thing someone can do on it? That gets you unstuck faster than the fix itself.
Stretch lane · finished early?
Pick ONE card from the sample-prompt menu, not five: ask for three takes, fix one mismatch, or apply a design schema.
"Show me 3 different takes on this screen, then I'll pick one."
"Make it look like airbnb.com: same feel, my content."
"I expected [X], it does [Y]. Fix that."
"Make it look like airbnb.com: same feel, my content."
"I expected [X], it does [Y]. Fix that."
One precise ask at a time. Checkpoint between asks.
30:00
First screens on the big screen
Win condition: the first screen of YOUR product renders. Destination: CLASS CHANNEL, URL or screenshot posted.
If you get stuck
Don't say "it's broken." Say this instead.
Gets you nowhere
it's not working. fix it??
Gets you unstuck
The page is blank when I open the URL. It should show the main screen: what a user sees when they arrive. Check what's wrong and fix it.
Stretch · this is not a toy loop
Every real company ships through the same five steps. You just ran the kid-size version.
Spec
your one-pager
Pull request
the agent's changes
Review
your teammates, before it ships
Deploy
the live URL you just posted
Rollback
commit before you experiment
Same five steps, every real engineering team, every day. You just ran all five.
Site walkthrough · giants are born ugly
Two more that started out looking like nothing.
Amazon is the real 1995 page, click to load it live. Google is verified via versionmuseum.com. Ugly first, huge later. Every time.
Go deeper · on your own time
Four places to keep exploring this
Lovable templates
A gallery of real, simple apps other people built with AI, browse for ideas.
lovable.dev/templates
Amazon & Google's ugly first versions
Real, restored screenshots of the giants at your exact starting point.
versionmuseum.com
"How we Claude Code"
For the ambitious: watch Anthropic's own engineers use the tool day to day.
youtube.com/@claude
Kimi Code docs
The official overview page. Day 3 goes deeper, this is a preview.
code.claude.com/docs
Next: 15:15
Team studio: the three tests set your build order, then you build. Standup at 16:15.
Every team's screen goes up at standup, whatever you have by then. Then tonight, you leave the building: real interviews with real strangers, starting with your own family. Tomorrow 9:00, the best quotes get read aloud.