Before the startup

Source: Lecture 3 - Before the Startup (Paul Graham), YC Root Access, 48:07, uploaded 2014-09-30, playlist index 231.

Paul Graham opens by asking what he would tell his own children about startups. His two-year-old says he will become a bat when asked what comes after two. Graham prefers the answer because it is more interesting, and then uses the joke to set the level of the lecture: the advice is aimed at young people, yet it comes from someone who has spent years watching founders make the same mistakes.

Startups against ordinary intuition

Graham compares starting a startup with learning to ski. A beginner who wants to slow down leans back and accelerates out of control. Skiing requires a small set of counterintuitive habits that eventually become automatic. Startups have the same problem. A founder’s ordinary instincts can lead in the wrong direction, so the first useful piece of knowledge is the knowledge that instinct may fail.

This explains why YC founders often ignore advice that later seems obvious. Graham describes the YC partners’ function as telling founders things they will disregard, then hearing a year later that the founders wish they had listened. Advice that agrees with an existing intuition feels unnecessary. A ski instructor is useful because skiing is strange; a running instructor would be a much harder business to justify. Graham describes YC as a business ski instructor for people travelling uphill.

People are an exception. A life spent meeting people gives a founder some relevant experience, even though it gives little direct preparation for starting a company. Graham says founders should trust their misgivings when an impressive person feels wrong. Business can distort this judgement for engineers who assume that a certain amount of unpleasantness must be normal. He recommends choosing co-founders, employees, and business partners with the same care used to choose friends, after enough time has passed for opposed interests to reveal character.

Users are the test

The second point is that startup success requires knowledge of the users rather than expertise in startups. Graham uses Mark Zuckerberg as the example. Zuckerberg did not know how to start a company. Facebook was first incorporated as a Florida LLC, which Graham treats as evidence of beginner status, and Zuckerberg may have paid little attention to the mechanics of the angel round that brought in Ron Conway’s cheque. Graham’s claim is that Facebook worked because Zuckerberg understood its users.

Young founders often imitate the outward forms of a startup. They develop a plausible idea, raise money at a good valuation, rent an office in SoMa, and hire friends. Then they discover that the activity has produced an office and a cap table without producing something people want. Graham and his partners called this “playing house”. The phrase captures the substitution at work: the founder performs recognisable startup actions while postponing contact with the users who decide whether the product has value.

Graham links this habit to education. Students spend years completing artificial tasks whose measurements can be gamed. He describes studying for exams by predicting the small set of ideas that could become exam questions, then preparing those answers in advance. In Automata Theory, he says, there were only a few sensible questions to ask, so the test became a test of which questions appeared. That method can produce good grades while leaving the subject itself partly untouched.

Fundraising becomes the imagined measure of startup success because it looks like the next exam. Founders ask for the trick that will convince investors, then the trick that will produce growth. Graham’s answer is plain: a startup that grows quickly because users love the product is easy to explain to investors. He treats “growth hacks” with suspicion because the phrase encourages founders to search for a technique outside the product. The growth follows from making something people want and telling them that it exists.

The distinction matters because investors can sometimes be persuaded for a round or two. Users have a simpler test. They care whether the software does what they need. A founder can create the appearance of progress long enough to raise more money, yet the equity-funded performance only consumes the founder’s own time if usage remains flat. The real problem stays in the product.

That is where gaming the system stops working. A large company may reward someone who flatters the right manager or sends emails late at night to look productive. A startup has fewer layers between the work and the user. The founder can mislead an investor for a while, though the user still decides whether the software earns a place in their life. Graham presents this as unusually good news about work: some parts of the world still make it hard to win through appearances.

The cost of making a startup your life

The fourth point is the scale of the commitment. A startup takes over its founders’ lives for years if it succeeds. Graham asks the audience to imagine Larry Page beginning to run as fast as he could at 25 and never stopping to catch his breath. Each day brings a problem inside Google that only its central leader can resolve. A week away creates a backlog, while the public grants a billionaire little sympathy for complaining about the burden.

Success hides this burden from the people who achieve it. Graham compares a founder to an Olympic sprinter whose effort stays invisible after the race. YC had funded several companies that counted as major successes, and the founders gave the same report: the problems change, yet the amount of worry does not fall. Construction delays in a new London office replace a broken air conditioner in a studio flat. The problem becomes more expensive and more public while remaining a problem that someone has to carry.

He compares a successful startup with having children because both choices alter a life for good. The comparison leads to a warning about timing. Universities supply contraception while also building entrepreneurship programmes and startup incubators, even though a startup can consume the attention that a student needs for other parts of life. Graham’s answer is that a university can teach about startups in the way a linguistics class teaches about languages. It cannot teach the particular users whose needs a founder must understand, and that knowledge appears through doing the work.

For that reason, Graham tells the students not to start a startup in college. He frames the advice as something he would give his own children. The early twenties offer forms of exploration that become harder once a company has taken over: a project with no clear payoff, cheap travel without a fixed return date, and the chance to follow an interest before it has to justify itself. Mark Zuckerberg can charter a jet, yet Facebook has removed the possibility of backpacking around Thailand with no public role and no schedule. The company runs him as much as he runs it.

Graham allows one exception. A side project can take off on its own, as Facebook did, and then the founder has a real decision to make. Most startups require their founders to force that takeoff. He calls it needlessly foolish to make the choice at 20 when waiting raises the chance of success and preserves years of open-ended exploration.

Aptitude is unavailable in advance

Starting a startup is hard enough that Graham asks how a person can know whether they are capable of it. His fifth point is that they cannot know in advance. A person’s past may reveal something about their prospects as a mathematician or a professional athlete, since those activities resemble earlier tests. Starting a startup changes the person doing it, so the estimate concerns a future self as much as a present one.

Graham had spent nine years trying to predict which YC founders would become tough and ambitious. He says intelligence was easy to estimate in a short conversation, much as a tennis player can reveal whether they return a ball. Ambition and resilience were different. He learned to keep an open mind about which companies in a batch would become stars. Confident founders who had succeeded at previous artificial tests performed no better by default than people who arrived convinced that their admission was a mistake. He recalls the same pattern from military recruits, where swagger does not predict toughness.

Fear still contains information. Someone who feels completely terrified of starting a startup probably should choose another path, unless fear itself attracts them. Uncertainty carries less information. The only way to discover what the work will make of an uncertain person is to try it, with Graham adding the timing qualification that the attempt belongs later in life rather than automatically in college.

Ideas that arrive sideways

The final point concerns the two things a startup needs at the beginning: an idea and co-founders. Graham says the usual method for finding both is the same. A person should avoid trying to invent a startup idea on command because deliberate brainstorming tends to produce ideas that sound plausible while remaining bad. Their plausibility wastes time because it fools the founder and the people hearing the pitch.

The better method is to develop a mind that notices ideas without searching for them. Yahoo, Google, Facebook, and Apple began as side projects rather than as companies assembled from an explicit startup plan. The strongest ideas often begin as outliers that the conscious mind would reject as businesses. The project comes first, and the company follows when the project starts demanding a larger share of life.

Graham reduces the preparation to three conditions: learn about things that matter, work on problems that interest you, and work with people you like and respect. The last condition produces co-founders alongside ideas. He first wrote a narrower version that advised people to become good at a technology, then widened it after considering Brian Chesky and Joe Gebbia. The Airbnb founders went to art school. Their advantage came through design and through organising people well enough to make a project happen.

The problem has to stretch the person working on it, though Graham cannot give a general test for what counts as important. Twitch began as Justin.tv, a service for broadcasting people playing video games. Graham initially thought the idea was absurd, and it became a good business. His own interest serves as the only guide he can offer. He repeatedly worked on subjects because they held his attention, and they later became useful in worldly ways. He says YC itself began as something interesting rather than as a planned route to a startup institution.

Technology supplies one more practical route. Graham describes it as a spreading edge, with each point on the edge presenting a problem that has recently become possible to solve. Paul Buchheit calls the condition living in the future. A person close to the edge sees a new product as an obvious absence while other people see an implausible prediction.

He gives the example of a Harvard graduate student in the mid-1990s who wrote voice-over-IP software so he could speak with his girlfriend in Taiwan without paying long-distance charges. Network expertise made it natural to turn sound into packets and send them across the internet. The student never meant to start a company. The problem was real to him, the technical solution was within reach, and the startup-shaped idea appeared as a side effect of solving his own problem.

That is why Graham recommends the classic idea of college as education for its own sake to students who may become founders. The part of entrepreneurship that matters most is domain knowledge. Larry Page became Larry Page because he understood search, and his understanding grew from interest rather than from a plan to use search as a startup advantage. Graham’s final advice for young founders is two words: “just learn”.

Questions after the lecture

The questions expose the edges of the advice. A non-technical founder can supply domain knowledge and handle the work around a product. In Graham’s Uber example, someone who understands the limo business could recruit drivers and manage the non-software operation, while the technical founder writes the iPhone and Android apps. In a pure technology startup, he jokes, the non-technical founder does sales and brings coffee and cheeseburgers. The serious point is that the useful contribution comes from knowing the domain and its users.

He is sceptical of business school for aspiring founders because management becomes a large problem after a startup succeeds. Early work centres on developing a product. Design school may be closer to that need, although Graham says that doing the work teaches it faster than a course. He revises an earlier piece of advice in the same way: he once told would-be founders to work at another company for several years, then decided that trying to start a company teaches the relevant lessons faster even when the attempt fails.

The first employees should resemble founders. They need enough self-motivation to work as peers rather than as people who require close management. Graham makes room for an exception when a company needs a specialist with rare technical knowledge. Such a person may need more support with ordinary life than with the specialised problem, and the founders may still decide that the knowledge is worth the arrangement. The rule remains a preference for people who carry their own work.

When asked whether the market is in a bubble, Graham distinguishes high prices from a bubble in which investors knowingly buy poor companies in the hope of selling them to a greater fool. He identifies the late 1990s as that kind of market, drawing on his own experience there, and says that the 2014 market has high valuations without meeting his definition of a bubble. He still tells founders to assume that the next funding round may become harder, since prices can fall and recessions can arrive between rounds. This answer belongs to the market he saw in 2014 rather than to a current market assessment.

The audience also asks about startup labs that try to spin off many companies. Graham points to Twitter as a side project inside a podcasting company, then says the model may work when the people running it use their own money. He treats the company-within-a-company pattern as plausible because some important startups have emerged from a side project that outgrew its original purpose.

On female founders and funding, Graham says his experience suggests that women have a harder time raising money and refers to interviews that Jessica Livingston was preparing. His advice is to make the company perform so well that the growth becomes hard to ignore. He describes tweeting an unnamed female-founded company’s strong growth graph so that venture capitalists would ask who had made it. The graph carried no gender marker, which let the investors encounter the evidence before the founder’s identity. The observation is an anecdote and a tactic, not a study of the funding gap.

Asked what he would study in college, Graham says physics because it is the subject he missed. He catches himself before turning that preference into career advice. The point of following curiosity is that its usefulness often appears later. The same answer shapes his account of work. YC gave him deadlines and applications that forced him to work carefully. Essays arrive involuntarily while he walks down the street. He has systems for work that has to happen and attention for work that he cannot stop doing, rather than a general method for manufacturing efficiency.

A side project becomes a startup when it takes an alarming share of the founder’s life. A student who spends the whole day on a project and risks failing classes has reached that threshold. When a founder asks what to do with a startup that grows slowly, Graham points to his essay “Do Things that Don’t Scale” and refuses to reconstruct it from memory. The answer remains consistent with the lecture: go closer to the users and solve the real problem before treating growth as a separate trick.

Graham cannot think of a class of startup that YC could not help, apart from a company that will fail or founders who are intolerable to work with. He says many startup problems recur across domains, which is why YC focuses on them. When asked how to identify an important problem, he admits that “work on interesting things” begs the question. His only partial test is personal: people with a taste for interesting problems tend to find known boring work intolerable. He also admits that this test confuses taste with self-discipline and leaves the larger question open.

The final question returns to the advice to work with people one likes. A group built from friends can develop a monoculture and miss some problems. Graham accepts the blind spot and judges the cost smaller than the benefits of working with people whose character is already known. He says successful startups often begin with friends from college. The lecture ends with that empirical claim still exposed as a trade-off rather than solved by a general hiring formula.

Limits

The lecture records Graham’s advice and YC experience in 2014. His examples of Facebook, Airbnb, Twitch, Twitter, Google, Apple, and YC are illustrative stories, while his statements about founder toughness, female founders, hiring, and funding markets arrive without a study, sample, or method that can be checked from the video. The market answer is historical. The comments on gender and growth rely on his observations and a reported set of interviews. The caption track also mishears some proper names and phrases, so this note keeps the examples whose meaning is clear and treats the speaker’s generalisations as claims from the talk.

Further reading / references

• Annotated lecture transcript, linked in the video description.
• Lecture 3 readings, linked in the video description.
• Course discussion for Lecture 3, linked in the video description.

Related: startup timing, startup funding, venture funding and control, value proposition design.