ISB AI Founders
Day 3 afternoon · 14:05–15:30

Feature blitz. Strangers are coming.

ISB AI FoundersBUILD
You first

Of the 5 strangers you're about to hand your product to, how many will use it exactly the way you imagined?

Tap to reveal
Basically none.
That's not a design failure, that's the entire reason a gauntlet exists. You already know your product. In 90 minutes, strangers do not.
Act 1 · 14:05–14:45 · Feature blitz

Freeze by 14:45. Strangers arrive the moment you do.

No fix buffer this time. Ship it, deploy it, freeze it, hand it to a stranger.
40:00 on the clock
YOUR TOOLS· Build: Kimi Code (every laptop, no VPN) · v0 (school device backup) · Same prompts, same rules
Win condition
The public URL loads on a phone that is not yours.
The core action works start to finish.
Your S8 style direction is visible on the page.
Ship checklist
1. Deploy.
2. Test on a teammate's phone.
3. Checkpoint commit.
Roles, all three running
Driver builds.
Navigator tests on their own phone.
Third teammate preps the observation sheets for 14:45.
Freeze by 14:45. No fix buffer this time. Strangers walk in the moment the timer hits zero.
40:00
One timer, the whole build
Between acts · the one rule

Silent observation: greet them, hand it over, say nothing else.

You may not
Explain.
Steer.
"Wait, click there."
You may
Greet them.
Hand over the phone or laptop.
Say nothing else.
Every instinct to help is a bug report. Write it down instead of fixing it out loud.
Act 2 · 14:45–15:30 · The stranger gauntlet · runs itself

5 cold users. The rotation runs the room. You write down every word.

45:00 on the clock
Rotation, posted
Team A tests B. B tests C.
C tests D. D tests A.
Then shift by two.
Who's cold
Other teams' members, staff, anyone who's never seen your product.
Pacing, self-timed
~9 min per user, 5 users. The clock paces the round, nobody calls time.
Swap: one observer per user. Swap after each round, everyone observes at least once.
Log: what they did, what they expected, what happened, does it block the core action?
The board runs this hour. The confusion log does not care who is watching.
45:00
5 users, silent the whole time
The confusion log format

One row per user. Five columns, no more.

User # What they did What they expected What happened Blocks core action?
1.........Y / N
2.........Y / N
Exact words, not your summary. "I don't know what this button does" beats "confusing button."
When it breaks live (and it will)

Don't apologize. Note it. Reset. Next user.

A crash in front of a stranger on day 3 is the cheapest lesson you will ever buy.
15:30 · 10 minutes

Break.

Water, bathroom, stretch. Keep your confusion log with you. S14 Multimedia AI starts the moment you're back at 15:40.
15:40 · Next up

S14, Multimedia AI. Keep your confusion log close.

You just watched 5 strangers use your priced product in silence. That evidence doesn't get filed, it gets clustered live at 16:15 standup: fastest fix, capped at five, no wall time needed.
Before you leave
LEAVE WITH · your app frozen and gauntlet-tested, plus a confusion log with exact words.
DECIDED · Nothing yet. Cluster and cap it at five, fast and live, at 16:15 standup.
BRING NEXT · Tonight, 2 people at home use your URL while you watch in silence. Log it. Tomorrow S9 turns both logs into the fix list you actually build.
16:15 · Standup

Cluster your log fast. Say your top fix out loud.

You watched strangers use your priced product for the first time. That's the whole job. Tonight, two more strangers get the same treatment. Tomorrow, S9 turns all of it into what you actually build next.