ISB AI Founders
Day 4 · 9:15

The product loop

ISB AI FoundersS9
You first

Pick an app you open every single day. When did it last change on you?

▸ click to reveal after they answer
Probably yesterday. You didn't notice, because it never throws a parade about it. That is the loop, working exactly as designed.
9:41
Updated last night. You didn't notice.
Loops, not lines

Real products are never finished. They are looped.

Ship
Watch
Measure
Fix
Every app on your phone runs this loop, forever. Today, so does yours.
The ladder, named

There isn't one loop this week. There are three.

Loop 1
BUILD
describe→run→test→gap
You've run this one since Day 1.
Loop 2
PRODUCT
ship→watch→measure→fix
Today. The whole session.
Loop 3
GROWTH
users→bring→users
What real companies chase next. Not today's job.
Exhibit A · every loop starts with a guess

Instagram's first version was called Burbn, and it was a mess.

V1 is a hypothesis, not a masterpiece. Two founders, Kevin Systrom and Mike Krieger, built a check-in app like Foursquare: points for hanging out with friends, plans you could make, badges, AND photo sharing bolted onto the side.
What the logs actually showed

Nobody checked in. Nobody made plans. Everybody posted photos.

The founders watched real usage, not opinions, and found one feature buried in a cluttered app doing all the work: a photo, a filter, a caption.
The cut

They deleted the whole app and kept the one feature.

Camera, filter, post, comment. Renamed Instagram, relaunched October 2010: 25,000 signups on day one, over a million users within two months. That is a zoom, not a persevere. Keep that word.
Exhibit B · one more, for good measure

YouTube was supposed to be a dating site.

In 2005 the founders built "Tune In Hook Up": upload a video describing your type. Nobody dated. People just uploaded random videos of pets and parties, like this one, the actual first video ever posted to the site.
19 seconds, a guy, an elephant enclosure. The dating site never happened. This did. Tap to play · youtube.com/watch?v=jNQXAC9IVRw
Real people, real behavior

Watch what they do, not what they say.

Two teams, same testers

Which one is telling you the truth?

Opinions
"I love everything, keep it all, maybe add a dark mode too!"
Behavior
4 of 5 testers opened the calendar once and never touched it again. All 5 used the checklist every single day.
You first

Which of these three calls do you think is hardest to make honestly?

▸ click to reveal after they answer
Persevere. It's the one call that lets you avoid admitting anything has to change. Zoom and pivot both force you to cut something you built. Persevere feels safe, which is exactly why it needs the most evidence.
The decision

Evidence, not attachment

The evidence says...
PERSEVERE
The evidence earns it. Show the exact line.
ZOOM
Rebuild around the one thing users grabbed.
PIVOT
Problem #2, carrying all your new skills.
Evidence, not attachment

Same fork. Three teams. Three different calls.

PERSEVERE · StudyBuddy
Gauntlet log: 4 of 5 strangers understood the flashcard app in under 10 seconds. 3 asked to keep using it after the test.
Call: persevere. Go deeper on what already works.
ZOOM · ChoreQuest
Points, a rewards store, a calendar, a leaderboard. Every tester ignored all of it except the daily checklist and streak counter.
Call: zoom. Cut the store and calendar, keep the checklist and streak.
PIVOT · CafeteriaFix
A lunch pre-order app to kill the line. Testers shrugged at pre-ordering but could not stop using the food-rating feature added as an afterthought.
Call: pivot. The real product is a review app, half already built.
Stretch · why good teams cut things they love

Hours spent is not a reason to keep it.

That is the sunk-cost trap: the more time you have already put into a feature, the harder it feels to cut, even when it is not working. The hours are spent either way. The only real question is whether it earns its place going forward. Google shut down Reader in 2013, a tool with a small but devoted following, because it no longer fit where the company was going. They killed it anyway.
Not all confusion is equal

Rank every fix by whether it blocks the core action.

Your core action is the one thing your product must let someone do. A confusion that stops that is a 5. A typo in the footer is a 1. Rank first, fix in that order.
Team ChoreQuest's core action was "mark a chore done." "I couldn't find the checkbox" blocks it: rank that a 5. "The owl mascot's arms look weird" is real feedback and irrelevant to Saturday: rank it last, or not at all. Same test as Day 1: does the fix make the product more of a painkiller, used again tomorrow without a reminder, or does it stay a vitamin?
Copy this prompt · run it now

The fix-list run

Here are observation notes from users trying my product. Cluster the confusions into patterns, rank them by how much they block the core action, and give me a fix list of at most 5 items. For each fix, also flag: does it make the product more of a painkiller, used again tomorrow without a reminder, or does it stay a vitamin? Then tell me: does this evidence say persevere, zoom in on one feature, or pivot? (Day 2 you graded the PROBLEM with REAL/WEAK/DEAD. Today you grade the PRODUCT. Same honesty.)
Watch it once, live

I run the loop on one real note. Then it's yours.

LIVE DEMO · no rehearsal
"Cluster these confusions. Give me a fix list of at most 5."
One real line from a real gauntlet log, ranked live, arguments included.
YOU DO
Have your own gauntlet log and last night's homework notes open and ready. Watch for the shape of a good fix list: capped at five, ranked by what blocks the core action. You run this exact move next.
Exercise 1 · do it now
FOLLOW ALONG · laptops open

Start the fix list

7:00 on the clock
1Paste gauntlet log + last night's home observations.
2Get the ranked fix list. Cap it at five. This starts the list; the 10:00 studio finishes and locks it.
3Disagree with Claude? Good. Defend it with a quote.
7:00
Then: into the 10:00 studio
Give it a memory, fast

Opinions lie. A counter never does.

Give your product a memory: three numbers, nothing more. The core action, completed. Where people quit before finishing it. Who comes back a second time. Resist tracking everything just because you can.
Copy this prompt · run it in Kimi Code

A few minutes to your first real number

Add a simple counter to my app: every time a user completes [my core action], log it with a timestamp to a small file or table. Show me the running total somewhere I can check any time, like a console log or a tiny admin page. Do not track anything else yet.
You first · I'm calling on someone

Would you rather have 50 strangers try your app once, or 5 who came back today on their own?

▸ click to reveal after they answer
The 5. Y Combinator gave Airbnb's founders exactly this advice, early on: better to have 100 people who love you than a million who kind of like you. Your version this week is smaller, same math. 50 one-tap strangers are worth nothing tomorrow. 5 who opened it again are your first real users.
50
clicked once, gone
5
came back today
Before you declare

Name the number before you look.

Pick the threshold now, before you check the log. For example: if fewer than 3 of 5 strangers complete the core action, we zoom. Decide it in advance, or you will bend the number to match whatever you already want to believe.
Exercise 2 · say it out loud
FOLLOW ALONG · laptops open

Make the call

5:00 on the clock
Not a group vote: your team's honest read of the evidence, said out loud to the room. Once you've said it, it's real.
1PERSEVERE, ZOOM, or PIVOT: one sentence. (Team split? Your Day 1 contract's decision rule was written for exactly this moment.)
2Attach the single evidence line that earns it, and the number you named in advance.
3Say it to the room. Anyone can challenge one assumption behind your call, out loud.
4Win condition: your call survives the challenge without changing your evidence.
5:00
Say it. Defend it.
The unpopular truth

The team that pivots today is braver than the team that perseveres out of habit.

A pivot means you built real evidence, read it honestly, defended it against a challenge, and had the guts to act on it with hours on the clock. That deserves louder applause than persevere ever will.
Checkpoint

A fix list started, and a call made, before the studio takes over.

Two artifacts, both required: a fix list started, at most 5, ranked by user pain, and a persevere, zoom, or pivot call, said out loud with one evidence line. Backed by a quote, not a feeling. The 10:00 studio locks both in before the build block runs.
“Do things that don't scale. Charge for it.” Today you proved it.
Exit ticket
Leave with
A fix list started, five items max, ordered by user pain.
Decided
Persevere, zoom, or pivot. Said out loud, with the evidence line that earned it. It survived the room's challenge.
Bring next
10:00: the team studio finishes and locks the fix list and the call. 10:25: break. 10:35: the build block executes it. S10, the launch video, is at 13:15.