ISB AI Founders
Day 1 · 10:45

Brainstorming problems

ISB AI FoundersS2
You first

Who has a great app idea?

▸ click to reveal after they answer
Wrong question. Nobody needs your idea. Somebody has a problem. We hunt problems today, not ideas.
Ideas vs problems

Nobody needs your idea. Somebody has a problem.

An idea: an app for studying. A problem: IB students lose their past-paper notes across five different apps the night before mocks. One of these has a person in it.
Idea vs problem · gaming

Same topic, totally different value

Just an idea
an app for gamers
An actual problem
day students get dropped from ranked queue when the dorm wifi cuts out at 9pm study hall, so they switch to mobile data and burn through it in a week
Idea vs problem · group chats

Same topic, totally different value

Just an idea
an app to organize group chats
An actual problem
new Y10 students don't know which of the six class WeChat groups actually matters, so they mute all of them and miss the football signup
Where big ideas actually start

Somebody, somewhere, was just annoyed.

Started as an annoyance · 2007

Airbnb: no rent money, a sold-out conference, and three air mattresses

Two guys couldn't afford rent in San Francisco. A design conference sold out every hotel in the city. They put three air mattresses on their floor and charged strangers to sleep on them. That was the whole company, before it was a company.
Started as an annoyance · 2006

Dropbox: a forgotten USB stick on a bus

Drew Houston kept forgetting his USB drive and losing hours of work between machines. On one bus ride it happened again, and he started writing the fix for himself, on the spot. That personal annoyance became a company used by hundreds of millions of people.
Started as an annoyance · 2004

Yelp: sick, alone, and no way to find a doctor

One of the founders caught the flu in a new city and had no idea which local doctor was any good. The internet had nothing useful to say. That gap became a review site used by hundreds of millions of people.
The format

Every thesis has this exact shape

[person]
Person
struggles to
[job]
Job
when
[moment]
Moment
, so they
[bad workaround]
Workaround
.
Four brackets. If you cannot fill all four, you do not have a problem yet. And a real thesis is falsifiable: name the person, name the moment it bites, name what they do about it today, and a stranger could go check whether you're wrong.
This exact shape, before it was worth anything

Twitter's co-founder sketched his thesis on notebook paper.

[person] struggles with [job]... at [the moment]... so they [bad workaround]
Same four brackets you just learned. Dorsey's real notebook page, posted by him on flickr, still online today.
flickr.com/photos/jackdorsey/182613360
Bracket by bracket

Four brackets, no exceptions

[person]
Not "students." Try "IB students studying for mocks." Point-at-able.
[job]
Not the app. "Studying" isn't a job. "Finding last year's paper at 11pm" is.
[moment]
Not "sometimes." A time, a place, a deadline breathing on it.
[bad workaround]
What they do instead, badly. A screenshot folder is a receipt for real pain.
All four brackets, one sentence

IB students struggle to find last year's practice paper when it's 11pm the night before mocks, so they screenshot everything into one giant camera roll.

Read it as one breath. If a stranger can picture the exact person and moment, it works.
Vague dies, specific lives

The bar

Dead on arrival
an app that helps students study better
Alive
IB students lose their past-paper notes across five different apps the night before mocks, so they screenshot everything into one giant camera roll
Vague dies, specific lives

Same test, new topic

Dead on arrival
an app for making friends
Alive
new boarding students eat lunch alone for the first two weeks because table seating is unspoken and nobody explains it, so they eat in the library instead
Finding the idea · Paul Graham, Y Combinator

What makes a problem worth your week?

rare + painful
Real, but only once in a while. A note, not a company.
weekly + painful · the painkiller corner
The gold corner. This is what you're hunting for.
rare + mild
The trap corner. "Interesting" ideas nobody actually needs.
weekly + mild · the vitamin corner
Nice to have. Forgotten by Tuesday.
how often it bites: rareweekly
Paul Graham, Y Combinator co-founder: the best startup ideas are problems the founders themselves have. That gold corner is also the classic painkiller test: a painkiller gets sought out right now, a vitamin gets forgotten by Tuesday.
Dig a well, not a crater · Paul Graham

A few people who want it badly beats a lot who want it a little

The crater
wide and shallow. a large number of people who want it a little.
The well
narrow and deep. a small number of people who want it desperately.
Start painfully small · the wedge

Does it have to be big? No.

Amazon
sold only books
→
everything you can buy online
Facebook
one college campus
→
billions of people
The giants started as embarrassingly narrow wedges. A small wedge you can win beats a big vision you can't start. Your one-sentence thesis, one person, one moment, IS the wedge.
Tar-pit ideas · Y Combinator

Ideas that look great and trap thousands of founders

Classic tar pit
an app to figure out what to do with your friends tonight
Classic tar pit
a discovery app for restaurants, music, or clothes
Classic tar pit
a social network for [any specific group of people]
Thousands of founders have tried all three. Why they trap: the problem is real but mild, and the fix needs everyone to change habits at the same time. Structurally brutal, no matter how good your app is.
You first · I'm calling on someone

Good, or tar pit?

(a) an app that tells you what to do this weekend with friends
▸ reveal
TAR PIT. Classic tar pit, mild problem everyone forgets by Tuesday.
(b) IB students can't find last year's practice papers at 11pm before mocks
▸ reveal
GOOD. Weekly, painful, a specific person: this is the thesis format.
(c) a social network for students at international schools
▸ reveal
TAR PIT. Needs everyone to switch apps at once.
(d) the cafeteria line makes you late to class every single day
▸ reveal
GOOD. Daily, painful, and you have it yourself.
You first · I'm calling on someone

Would you rather 100 people love your thing, or a million kind of like it?

▸ click to reveal after they answer
100 people who love you beats a million who kind of like you. Brian Chesky's rule, Y Combinator canon. This week's version: five real users who love it by Sunday.
Checkpoint · why you

The founder who already has the problem wins.

Airbnb, Dropbox, Yelp, all of it: same founder, same problem. Your annoyance list is not a warm-up. It's the well. Time to start digging.
Block 1 · fill the wall
OFFLINE · screens closed

Three stages, one block. Two solo, then the table collides.

Stage 1 · min 8
SOLO · diverge
Annoyance inventory
Write 8 things that annoyed you in the last 7 days. Small counts: the queue, the group chat, the charger. Nothing is too small.
Stage 2 · circle 5+
SOLO · diverge
Day replay
Map yesterday hour by hour. Circle at least 5 moments of friction. Star anything that happens every single day.
Stage 3 · min 5
TABLE · collide
Trade & tally
Trade lists around your table. Tally a mark on anything you also have. 3+ tallies = a star: that is not your annoyance anymore, it is a market. Then post to the wall together, one note per item, min 5 per person, big letters readable from 2 meters, no sorting yet.
24:00
Self-paced. Then: cluster as a room
Block 1 · sharpen as you go

Two things to keep in your head

What counts
Small stuff counts: the wifi drop, the queue, the dead charger. Don't wait for a "big" idea. You'll filter later, not now.
Already done? Harder round
For each starred annoyance, name a specific other person who has it too. Not "everyone." Your roommate, the kid two seats over. Can't name anyone? It might just be yours, and it dies at the wall.
Block 1 · debrief
OFFLINE · screens closed

Which annoyance got the most tallies at your table?

A 3+ tally star is not a hunch anymore, it is a market: other people already have this. Frequency beats intensity, the starred everyday moments beat the dramatic once-a-year ones. If the wall looks full, you're ready.
Whole-room · cluster
OFFLINE · screens closed

Cluster into territories

4:00 on the clock
1As a room, physically move notes into groups that feel related.
2Give each cluster a rough name: food, wifi, group chats, sports, family.
3Nobody owns a cluster yet. Just organize the mess.
4:00
Then: dot-vote the hottest ones
Whole-room · dot-vote
OFFLINE · screens closed

Dot-vote the hottest clusters

no clock, just move
1Everyone gets 3 dot stickers.
2Vote for the clusters that feel most alive, not your own note.
3A heat map appears on the wall in under a minute.
Then: THE WALL is done, read it.
Whole-room · claim
OFFLINE · screens closed

Claim one problem from a hot cluster

Pick a note, yours or someone else's, from a cluster with real heat. You do not have to invent alone. The wall just did research for you.
Exercise · the ridiculous ten
OFFLINE · sticky notes

Ten solutions. Five minutes. The dumber the better.

1Your claimed problem. One sticky per solution. Ten minimum, more is better.
2Ridiculous is welcome: a robot, a rumor, a rule change, a bake sale. Wild ideas hide real ones.
3No judging, no "that won't work." Quantity wins. Then read your wildest one aloud at the table.
Stick them all under your problem on the wall. This afternoon the AI storms solutions with you; right now your brain goes first.
5:00
10+ stickies under your flag
You first

Would a real person pay ¥10 for this to disappear? Prove it, don't guess.

▸ click to reveal after they answer
"Yes" is not proof. Proof is a behavior they already do: money already spent on a bad fix, time already burned on a workaround, an app already downloaded and abandoned. No existing behavior, no thesis yet. That's the ¥10 filter, adversarial version.
Block 2 · pressure-test
FOLLOW ALONG · laptops open

Four moves. Screens open, then paper, then your table.

1Ask the AI to evaluate it and improve it. Feed it your claimed problem in one prompt. Get back the attack on its weakest part, who already solves it and why they haven't killed it, and a sharper version of the problem, all in one run.
2The ¥10 filter. Would a real person pay ¥10 for this to disappear? Proof is a behavior they already do, not a guess.
3Write your thesis, then kill it yourself. One sentence, four brackets: person, job, moment, workaround. Print your name under it. Then thirty seconds: argue the strongest reason it's wrong. Fix the bracket that breaks, then tape it up.
4TABLE · rehearse. Read your thesis to your table. The table finds ONE hole before the room gets to. Patch it or defend it.
Before it's on the wall
a. names a person and a moment
b. survives the ¥10 behavior test
c. names the strongest competitor and the gap
d. the AI attacked it, and your fix survived
22:00
Self-paced. Then: gallery gauntlet
Copy this · flip back anytime

The prompt for block 2

1 · evaluate it
Here is my claimed problem: [X]. Evaluate it like a tough startup mentor. 1) Attack the weakest part. 2) Who already solves this, which one is strongest, and why has it not killed the problem? 3) Role-play the person who actually has this problem for three exchanges. 4) Then give me a sharper version of the problem in one sentence.
Part 2 of the answer is easy: a list of competitors. The part that matters: name the single strongest one, then finish out loud, they haven't killed this problem because ___.
You first

The AI just named your strongest competitor. Do you believe it?

▸ click to reveal after they answer
Not automatically. Ask it why it picked that one, right in the same chat. Check the answer against what you actually know from being in the room. If "why hasn't this been killed" doesn't survive your own gut check, keep pushing back before you write your thesis.
The gauntlet

Everyone reads theirs aloud. The room's job: find the hole, kindly.

Hole 1 · person
Could this be literally anyone? "Students" is a hole. "IB students at 11pm before mocks" is not.
Hole 2 · moment
Could this happen "sometimes, whenever"? If there's no specific trigger, it's not a moment yet.
Hole 3 · workaround
Is the workaround made up? A real one is something you've actually watched someone do.
Checkpoint · gallery walk

Read it aloud. If nobody finds a hole, it survives.

This wall is where teams come from at 1:15.
“Make something people want.” Today you proved it.
Whole-room · plant your flag
OFFLINE · screens closed

Say it out loud. Then star it on the wall.

“The problem I want to spend my week on is ___, because ___.”
Key rule
It does not have to be your own thesis. Choosing someone else's surviving problem off the wall is a founder move, not a defeat.
Physical action
Put a star sticker, or a marker star, on the thesis you just claimed, wherever it lives on the wall.
5:00
By lunch the wall shows which problems have gravity. Those clusters are where teams form at 13:15.
Go deeper · optional

Three real places this all comes from

The essay this session is built on
long read
Paul Graham, one of Y Combinator's founders, on why nobody needs your idea, somebody has a problem. Same lesson as today, in his own words.
paulgraham.com/startupideas.html
A Y Combinator partner, live
~33 min watch
Real YC companies, told by the person who helped fund them: how their founders actually found the idea that worked. Around ~33 min, save it for tonight if you are hooked.
youtube.com/watch?v=Th8JoIan4dg
A billion-dollar thesis, handwritten
1 min look
Twitter's actual first sketch, on notebook paper, posted by co-founder Jack Dorsey himself. Yours today looked a lot like this.
flickr.com/photos/jackdorsey/182613360
All three are on the portal too. Nobody has to read them before lunch.
Exit ticket
LEAVE WITH · One falsifiable thesis that survived the gauntlet, on the wall
DECIDED · Which problem you are spending your week on: starred on the wall, said out loud.
BRING NEXT · That problem, memorized. At 13:15 you pitch it six times in speed-dating. No thesis, no dating.
Next up · 13:15

Co-founder dating

Lunch first. If your thesis did not survive, that is the process working: you get first pick of the surviving problems after lunch. This afternoon you find who you build with.