YC Startup School, for people building something small

Start from a problem, not a technology 02:11

Ideas that begin with a tool looking for a use produce problems nobody has. Begin with a specific problem you have seen people struggle with.

you want to fall in love with a problem

Watch for tar-pit ideas 03:19

Some ideas look easy and come up again and again, yet fail for a structural reason. Search for the idea and talk to people who tried it before you start.

it seems like it could be easily solved with a startup but it's an illusion

The best problems have no alternative at all 08:53

If people currently do nothing about the problem because nothing exists, the pain is real and they will put up with a rough first version.

if the alternative to your solution is literally nothing that's what a good problem looks like

Look for what just changed 09:26

A new technology, rule or habit is often what opens a gap that was not there before.

only recently become possible or only recently become necessary

If you cannot decide, launch it 31:39

No checklist settles whether an idea is good. Putting it in front of people does.

the only way to know for sure if your startup idea is good is to just launch it and find out

You cannot find the right idea by thinking 01:07

The perfect idea cannot be worked out in the abstract. Contact with customers is what tells you what to build.

You can only figure out what you should be working on by making contact with reality and getting feedback from customers.

Juggling ideas gives you bad data 02:46

Split attention means no idea gets enough effort to show whether it works, so you may drop a good one or keep a bad one.

you won't get good signal about whether what you're doing actually works

Adapt it With a day job you cannot burn every boat, but the point stands: stop dividing your effort across projects.

Know your user's day well enough to do their job 04:27

A test of depth: could you run your customer's business tomorrow, with its crises and costs?

if I dropped you into a cleaning business tomorrow, would you know how to run it

Going deep finds the better idea 09:31

The first pain points you hear are on the surface. Staying with one idea is how you reach the structural problem underneath.

going deep isn't primarily a process for validating the idea you started with. It's a way to find the better idea underneath.

Resilience matters more than credentials 02:11

Getting through the long early stretch of rejection is the quality that predicts who keeps going.

it takes a lot of resilience to make it through this initial struggle to get your early users

Confidence is a poor sign of resilience 02:45

Quiet founders often turn out to be the ones who last.

some of the quietest and least confident seeming founders at first turned out to be the most resilient founders

One passionate user beats a big waitlist 15:19

A crude prototype that one person loves tells you more than sign-ups for something that does not exist yet.

a single passionate user of a crude prototype that was hacked together in a weekend means so much more than a million sign-ups

Do not pitch in the interview 06:07

Describing your idea biases the answers. Your job is to listen and ask open follow-ups.

to not introduce your idea until maybe at the end of the call or maybe not at all

Ask about what they do now 08:51

Ask how they handle the problem today and what they already do to solve it. Watch them do it if you can.

what do you do to solve this problem for yourself today it's really important to understand what they're doing now

Users know problems, not solutions 11:06

When someone asks for a feature, dig for the reason behind it. The reason is the useful part.

users generally have good problems but also generally bad solutions

Keep your first users in a group 15:31

Put early interviewees in a WhatsApp or Slack group, show them each new version, and ship in response to what they say.

create a slack instance or a WhatsApp group for with their customers

Early adopters forgive a broken product 03:22

People willing to try a new product already expect it to be rough. Those who leave after one bug were never going to use it.

the people who are interested in talking to a startup are early adopters they're used to using products that don't work very well

Fast to build, few features, a small group of users 07:15

The first versions of successful products shared all three.

all of these products were fast to build they could get out of the market quickly second they all had very limited functionality

Set a deadline and write the spec down 14:31

A few weeks, written features, and no re-debating what was agreed.

if you write it down then you can just focus on building instead of continuously debating what should be built

Cut to what a desperate user needs 15:05

Go through every feature and ask whether someone truly desperate would need it to start.

ask yourself there's a truly desperate customer need that feature to start

A hundred who love it 16:05

Better a small group that loves the product than a large one that half likes it.

far better to have a hundred people love your product than a hundred thousand who kind of like it

The first prototype is for showing 04:35

Build something in days whose only job is to be shown to users. It does not have to fully work.

build a prototype as soon as possible with the singular focus to build something to show and demo to users

Do things that don't scale 11:49

Onboard users by hand and give personal support instead of building self-service flows.

avoid things like automatic self onboarding because that adds a lot of engineering

The 90/10 solution 12:57

Limit the product to one situation, user type, device or place, and push the rest past launch.

restrict the product to work on limited dimensions

Don't build for scale early 16:52

Copying a big company's setup (Kubernetes, many microservices) before you have users is how startups die.

okay that's the average technical founder the average startup dies

Tech debt is fine before people want it 22:57

Bugs and shortcuts matter less than getting to something people use.

tech debt is totally fine

Plan in a file first 04:59

Write the plan with the AI in a markdown file in the project, cut what you dislike, and work through it a section at a time.

Put that in a markdown file inside your project folder and keep referring back to it.

Reset rather than pile fixes on top 06:38

Start each feature from a clean state in git and throw away attempts that go wrong.

don't be afraid to get reset head hard if it's not working and just roll the dice again

High-level tests catch what the AI breaks 07:11

End-to-end tests that click through the app catch the unrelated changes AI tools make.

I prefer to keep these tests super high level.

Give the AI the docs 11:04

Download the documentation of the APIs you use into the project and tell the AI to read it.

download all of the documentation for a given set of APIs and put them in a subdirectory of my working folder

Context like an employee 04:03

Models do their best work when given the context you would give a new hire.

you need to provide models with as much context as you would provide an employee

Specs and tests define the work 04:37

People write the specification and the tests that define success, and agents write code until the tests pass.

humans write a spec and a set of tests that define success

Adapt it A solo maker can work this way, but should still read what the agent produced.

An uncomfortable AI bill 08:32

Spending a lot on AI can replace much bigger costs elsewhere.

you should be willing to run an uncomfortably high API bill

Adapt it Only up to a budget. A free product has no revenue to cover a large bill, so set a daily cap and measure what it replaces.

Launch now 01:12

Your beliefs about how people will use the product stay theoretical until someone uses it.

the answer is as soon as possible. Most likely, right now.

Launch more than once 02:15

If nobody notices, change something and launch again.

Airbnb launched three times before it started getting real users.

Say what, not why 05:25

Describe the product in one sentence: the name and what it does, with no backstory.

starting with “what” you do, not “why” you do it

Launch where you already take part 16:29

Post in communities you actually belong to, sincerely, after learning what they care about.

Launch in communities that you really actively interact with.

Press does not make growth 19:40

Coverage reaches some early adopters but does not keep users coming. Build your own list of supporters instead.

press cannot be relied upon for sustainable growth.

Recruit the first users by hand 01:07

Products do not take off by themselves. Writing more code is a common way of avoiding this work.

you have to manually recruit your customers

Most people are not early adopters 12:07

Expect to reach out to many people to find the few who are, and do not try to convert the rest.

most people are not early adopters

Users who do not pay are not proof 13:45

A customer who does not pay is not a customer.

if you don't charge your customers they are not a customer and you don't have a company

Adapt it A deliberately free product can skip charging, but then free users are not proof that anyone wants it. Retention has to be.

Don't start with search and ads 22:07

People who arrive through Google or ads are hard to get on a call, so you learn nothing from them.

you can't learn from users if the first thousand or first-handed customers are brought in through Google and Facebook

Adapt it If search is how people find you, keep it, but give every visitor a way to reach you so you can still talk to some of them.

Early users buy on trust in you 02:45

The first people use your product because they trust you, not because the product is good yet.

the people buying your product early are buying because they trust you as a founder, not just because of the quality of the product.

Go where they already complain 06:01

Find the Reddit threads, groups and forums where people complain about the problem, and show up as a person.

Your job is to find it and then show up there as a real person.

Keep messages short 09:52

Under 75 words, one clear ask, and read it aloud to catch lines that sound machine-written.

First, keep it under 75 words.

Network, then manual work, then tools 11:32

Users one to three come from people you know, four to ten from unscalable effort, and only after that do outreach tools help.

Customers 1 through 3 almost always come from your personal network.

Adapt it The order holds for any product. The outreach tools at the end may never be needed for a small consumer one.

Count real use, not opens 05:02

Define an active user by an action that shows they got value from the product.

the best action to pick is one that is really correlated with the user getting real value from your product

Match the period to the product 06:07

Daily for something meant to be used daily, weekly or monthly for others.

you want to pick a time period that matches what you intend for your product

The shape matters, not the height 11:41

A retention curve that flattens means some users stay. One that keeps falling means you are only replacing those who leave.

the shape of the curve is what matters not the absolute number

Don't widen the window to look better 14:30

Longer periods make curves flatter and higher by definition.

by definition your cohort curves are going to be flatter and higher the larger you pick your time period

Onboarding is the cheapest fix 24:58

Helping new users get to the useful part is often the easiest way to lift retention.

investing in this often is like the cheapest and easiest way to improve your performance of your cohorts

Totals hide people 00:08

Daily or monthly active counts lump everyone together and can rise even when nobody enjoys the product.

relying on aggregate user metrics instead of understanding how individual users use their product

One row per user, one dot per day 01:43

Put a dot for each day a user did the valuable action and circle the day they signed up.

Each row represents one individual user.

Look before you compute 06:28

Patterns jump out of the picture that you would never have thought to test for.

Your brain will start to notice these patterns in a way you would never have understood on your own.

It can be your only dashboard 12:18

Until you have hundreds of users, this chart plus retention curves is enough.

until you have hundreds of users, a scatter chart can be your only dashboard

Make the moments worth sharing easy to share 03:31

Find the moments a user has something to show off, and make sharing them one tap.

what are the shareable moments?

Organic loops outlast ads 04:37

Effort on sharing and network effects keeps paying. Ad spend stops working the day you stop paying.

Viral loops and network effects pay back forever.

Know where every user came from 06:48

Use a tracking parameter in links, or simply ask at sign-up where they heard of you.

you should just ask your users when they signed up uh where they heard about you

Find the magic moment 15:07

Compare users who stayed with those who left to find an early action that predicts staying, then steer new users to it.

if you added seven friends in your first 10 days of usage

Adapt it With few users the comparison is noisy, but a rough look at what retained users did first still helps.

Ignore vanity metrics 02:46

Numbers that feel good or look good on LinkedIn make you work hard on the wrong thing.

we call these vanity metrics things that make you feel good and things that you can brag about on LinkedIn

Find the real bottleneck 07:48

When the main number is stuck, look for the actual cause. It may be simpler than you think.

it wasn't a ux friction issue or an app issue

Spot fake progress 11:10

Conferences, technical milestones and features nobody asked for feel like work but do not move the number.

launching an Android app unless you're hearing clearly from your users that this is a burning pain point

Churn beats feature requests 15:32

If users are leaving and nobody new is arriving, that is the problem to work on.

that's your biggest problem go solve that

Weekly growth targets 18:53

YC measures growth week over week, with 5 to 7 percent as good.

five to seven percent week over week is growth is good and a ten percent week over week growth is exceptional

Venture-backed only This bar comes from venture math. A small product should track its own trend, not this target.

Ads need enormous traffic 08:58

Advertising works as the main model only for the very largest sites.

you should not use ads as your primary

Charging teaches you 16:48

Asking for money tells you who values the product and how much, even when people say no.

it turns out that charging is actually one of the most effective ways to learn a lot of really important things about your business

Adapt it A product you mean to keep free can skip this, but it gives up the clearest signal of demand.

Get the order of magnitude right 19:00

Early on, only a tenfold mistake in price matters.

if you're charging ten dollars for your product and your customers are willing to pay a hundred you should probably change your price

Raise until people push back 22:22

If everyone accepts the price at once, it is probably too low.

another way to find your value is to keep incrementally raising prices until you get pushback from users

Adapt it Without sales calls, test a new price on new sign-ups only.

Most founders undercharge 22:57

A low price is easy for a bigger rival to beat and can make people trust the product less.

most startups are actually under charging you almost certainly are