ISB AI Founders
Day 2 · 13:15

The Verdict

ISB AI FoundersS6
You first

When's the last time you only remembered the compliments?

▸ click to reveal after they answer
That's confirmation bias. It's not lying, it's just how your brain protects an idea it already loves, and today it's coming for your project.
The bias, applied to you

You do this with test grades too.

You remember the 94 and "great work!" in green pen. You forget the same teacher wrote "see me about #4" on three tests in a row. Same paper. Two very different memories, and you kept the one that felt better because it didn't feel like cheating, it felt like being right.
Same report card, two readings

Which student are you?

Wishful
Teacher wrote "see me" on three different tests this term. I'm basically acing this class.
Honest
Teacher flagged the exact same mistake three times and I never once went to see her. That's not a compliment I'm missing, that's a pattern I'm avoiding.
The skill nobody teaches

Read what the evidence says. Not what you hoped it would say.

Teams that grade their own homework honestly today build things people want by Saturday.
From theory to your own notes

Now read what's actually there.

Now raise the stakes

Your interview notes have the same trap. Most of what's in there is politeness, not evidence.

You just spent two days getting real people to talk to you. Evidence is what they do, describe doing, or already pay to fix. Politeness is what they say to be nice to the kid with the clipboard. Your job now: tell the two apart.
Signal vs noise · pair 1 of 3

The compliment

Politeness
"Oh my gosh, that's such a cool idea!" (said while still scrolling their phone)
Evidence
"I already have three group chats just for splitting rent and I still miss the deadline every month." (pulls up the chat to show you)
Signal vs noise · pair 2 of 3

The hypothetical

Politeness
"Yeah, I'd definitely use that." (has never once searched for a fix)
Evidence
"I actually paid for an app that did half of this last year. Quit after a week, too slow."
Signal vs noise · pair 3 of 3

The enthusiasm

Politeness
"Wow, you're so smart for thinking of this!" (changes the subject immediately after)
Evidence
"Wait, show me that again. When would I actually get the reminder?" (leans in, asks a real follow-up)
You first

Be honest: which column does most of your notebook look like right now?

▸ click to reveal after they answer
If you're not sure, that's the tell. Go find out before you write a verdict, not after.
The verdicts

Grade your problem

REAL
They described the pain before you did
WEAK
Polite interest, zero behaviour behind it
DEAD
Nobody had it. Congratulations: you just saved four days
A real company, at exactly this stage

Buffer tested a fake pricing page before writing any code.

People clicked the pay button on a product that did not exist. That click was a REAL verdict. Your interviews are the same test, one day earlier.
The verdicts, with the evidence

What a real verdict actually sounds like

Your evidence, sorted
REAL
They described the pain before you did, already doing something imperfect about it.
"Four of five already track this by hand, unprompted. One showed me the actual spreadsheet."
WEAK
Polite interest, zero behaviour behind it.
"Everyone said 'yeah that'd be useful,' but four of five have spent nothing, not a yuan, not a minute, fixing it."
DEAD, and a win
Nobody had this problem left to solve. You just saved four days.
"One laughed and said, 'that's literally just Notion.' No gap left to build into."
Why hand this to Claude

Everything you collected, funneled down to one sentence

15
raw interview notes
→
3
patterns
→
1
strongest quote
→
1 verdict
REAL, WEAK, or DEAD
Claude does the counting in thirty seconds. You spend your five minutes arguing with the answer.
Copy this prompt · run it now

Notes in, patterns out

Here are my raw notes from 5+ customer interviews about [problem]. 1) Cluster them into patterns. 2) Surface the single strongest quote. 3) Now be brutal: which of my conclusions am I reaching because I want them to be true? What does the evidence NOT support? (Same four-part anatomy you learned Day 1, doing heavier work.)
Watch this once, then it's yours

Notes in, patterns out, live

LIVE DEMO · no rehearsal
"Cluster these raw notes. Then: what does the evidence NOT support?"
Run live on a sample set, clusters and all, including the ones that get argued with.
YOU DO
Open your own team's raw interview notes.
Have the prompt ready to paste in, but don't run it yet.
Watch how I argue with a cluster before you argue with yours.
When I say go, it's your notes, your five minutes.
Exercise 1 · do it now
FOLLOW ALONG · laptops open

Feed Claude your notes

5:00 on the clock
1Paste every note, including last night's homework interviews.
2Read the clusters. Argue with anything that feels wrong.
3Copy the strongest quote somewhere safe: it goes in your deck on Friday.
5:00
Destination: TEAM FOLDER
Debrief · what to listen for

Listen for the team that goes quiet reading step three.

That's the sound of a group finding out their favorite idea has a hole in it. Let the silence sit for a second before you move the room on.
The bias-check beat

Make Claude argue against you. On purpose.

Step 3 of that prompt is the whole session in one sentence. If it can't find something you wanted to hear but wasn't actually said, you didn't ask hard enough. Run it again, harder.
You first

What's the bravest thing Claude could say back to your team right now?

▸ click to reveal after they answer
Something like: "You counted your friend's laugh as evidence. It isn't. Nobody in these notes asked for this on their own." If that stings a little, it's working.
Why write it down

Clusters and a quote don't ship anything by themselves. One sentence has to.

REAL, WEAK or DEAD, defended in a single paragraph, with the evidence you just found and nothing you wish were there.
Exercise 2 · do it now
FOLLOW ALONG · laptops open

Write the verdict

5:00 on the clock
1REAL, WEAK or DEAD, one paragraph of evidence, real quotes only.
2WEAK or DEAD? Pick your #2 problem NOW: the whole afternoon is still ahead. That is the system working.
5:00
Destination: THE WALL
Debrief · what to listen for

Listen for the word "probably."

"Probably REAL" means you're not done rereading the notes. A verdict is a sentence you'd defend in front of the room, not a guess you're hoping survives it.
Now say it to the room

One quote. Sixty seconds. No feelings.

The defense format

Defend your verdict in 60 seconds. With a quote, not a feeling.

One team member reads it out loud. The room gets one question. Feelings are inadmissible; a direct quote is the only thing that counts as proof.
Two defenses, same 60 seconds

Which one survives the questions?

The stall
"We think it's REAL because... I mean, we just feel like people would use it, you know? Like it makes sense to us?"
The ship
"REAL. Four of five already track this by hand, unprompted, and one showed us the actual spreadsheet. Direct quote: 'I redo this every Sunday and it takes an hour.' That's going straight into Friday's deck."
If your verdict is WEAK or DEAD

Re-picking is not failure. It's Day 1's plan working, three days early.

1Open your notes. Your #2 problem is already in there, you interviewed about more than one thing, on purpose.
2Run the same prompt. Same clustering prompt, five minutes, new problem. You already know how.
3Same afternoon. You still hit wireframing at 14:45 with everyone else. Nobody can tell unless you say so.
Checkpoint

One verdict per team, on the wall, defended with a quote.

The afternoon only builds what survived the evidence. Everything else was still worth four days.
“Talk to users.” Today you proved it.
Go deeper · optional

Two founders who tested before they built

The essay on doing unscalable things
short read
Paul Graham on why the best founders start by doing embarrassing, hands-on things to get real evidence, not comfortable guesses.
paulgraham.com/ds.html
The fake pricing page that built Buffer
short read
Buffer's own founder tells it: he tested a pricing page before writing one line of code. 120 signups, one real paying customer, in 3 days. Evidence, not opinions, exactly like today.
medium.com/@joelgascoigne/how-to-successfully-validate-your-idea-with-a-landing-page-mvp
Both are on the portal too.
Next up

From verdict to wireframe.

14:15, your spec gets hardened by what you just proved. 14:45, you draw every screen on paper, boxes and arrows, before a single line of code. 15:15, that paper wireframe becomes prototype v0, live in Claude Code.