September 2022I recently told applicants to Y Combinator that the best advice Icould give for getting in, per word, was Explain what you've learned from users.That tests a lot of things: whether you're paying attention tousers, how well you understand them, and even how much they needwhat you're making.Afterward I asked myself the same question. What have I learnedfrom YC's users, the startups we've funded?The first thing that came to mind was that most startups have thesame problems. No two have exactly the same problems, but it'ssurprising how much the problems remain the same, regardless ofwhat they're making. Once you've advised 100 startups all doingdifferent things, you rarely encounter problems you haven't seenbefore.This fact is one of the things that makes YC work. But I didn'tknow it when we started YC. I only had a few data points: our ownstartup, and those started by friends. It was a surprise to me howoften the same problems recur in different forms. Many later stageinvestors might never realize this, because later stage investorsmight not advise 100 startups in their whole career, but a YC partnerwill get this much experience in the first year or two.That's one advantage of funding large numbers of early stage companiesrather than smaller numbers of later-stage ones. You get a lot ofdata. Not just because you're looking at more companies, but alsobecause more goes wrong.But knowing (nearly) all the problems startups can encounter doesn'tmean that advising them can be automated, or reduced to a formula.There's no substitute for individual office hours with a YC partner.Each startup is unique, which means they have to be advisedby specific partners who know them well.[1]We learned that the hard way, in the notorious "batch that brokeYC" in the summer of 2012. Up till that point we treated the partnersas a pool. When a startup requested office hours, they got the nextavailable slot posted by any partner. That meant every partner hadto know every startup. This worked fine up to 60 startups, but whenthe batch grew to 80, everything broke. The founders probably didn'trealize anything was wrong, but the partners were confused andunhappy because halfway through the batch they still didn't knowall the companies yet.[2]At first I was puzzled. How could things be fine at 60 startups andbroken at 80? It was only a third more. Then I realized what hadhappened. We were using an O(n2) algorithm. So of course it blewup.The solution we adopted was the classic one in these situations.We sharded the batch into smaller groups of startups, each overseenby a dedicated group of partners. That fixed the problem, and hasworked fine ever since. But the batch that broke YC was a powerfuldemonstration of how individualized the process of advising startupshas to be.Another related surprise is how bad founders can be at realizingwhat their problems are. Founders will sometimes come in to talkabout some problem, and we'll discover another much bigger one inthe course of the conversation. For example (and this case is alltoo common), founders will come in to talk about the difficultiesthey're having raising money, and after digging into their situation,it turns out the reason is that the company is doing badly, andinvestors can tell. Or founders will come in worried that they stillhaven't cracked the problem of user acquisition, and the reason turns outto be that their product isn't good enough. There have been timeswhen I've asked "Would you use this yourself, if you hadn't builtit?" and the founders, on thinking about it, said "No." Well, there'sthe reason you're having trouble getting users.Often founders know what their problems are, but not their relativeimportance.[3]They'll come in to talk about three problemsthey're worrying about. One is of moderate importance, one doesn'tmatter at all, and one will kill the company if it isn't addressedimmediately. It's like watching one of those horror movies wherethe heroine is deeply upset that her boyfriend cheated on her, andonly mildly curious about the door that's mysteriously ajar. Youwant to say: never mind about your boyfriend, think about that door!Fortunately in office hours you can. So while startups still diewith some regularity, it's rarely because they wandered into a roomcontaining a murderer. The YC partners can warn them where themurderers are.Not that founders listen. That was another big surprise: how oftenfounders don't listen to us. A couple weeks ago I talked to a partnerwho had been working for YC for a couple batches and was startingto see the pattern. "They come back a year later," she said, "andsay 'We wish we'd listened to you.'"It took me a long time to figure out why founders don't listen. Atfirst I thought it was mere stubbornness. That's part of the reason,but another and probably more important reason is that so much aboutstartups is counterintuitive. And when you tell someone somethingcounterintuitive, what it sounds to them is wrong. So the reasonfounders don't listen to us is that they don't believe us. Atleast not till experience teaches them otherwise.[4]The reason startups are so counterintuitive is that they're sodifferent from most people's other experiences. No one knows whatit's like except those who've done it. Which is why YC partnersshould usually have been founders themselves. But strangely enough,the counterintuitiveness of startups turns out to be another of thethings that make YC work. If it weren't counterintuitive, founderswouldn't need our advice about how to do it.Focus is doubly important for early stage startups, because notonly do they have a hundred different problems, they don't haveanyone to work on them except the founders. If the founders focuson things that don't matter, there's no one focusing on the thingsthat do. So the essence of what happens at YC is to figure out whichproblems matter most, then cook up ideas for solving them — ideallyat a resolution of a week or less — and then try those ideas andmeasure how well they worked. The focus is on action, with measurable,near-term results.This doesn't imply that founders should rush forward regardless ofthe consequences. If you correct course at a high enough frequency,you can be simultaneously decisive at a micro scale and tentativeat a macro scale. The result is a somewhat winding path, but executedvery rapidly, like the path a running back takes downfield. And inpractice there's less backtracking than you might expect. Foundersusually guess right about which direction to run in, especially ifthey have someone experienced like a YC partner to bounce theirhypotheses off. And when they guess wrong, they notice fast, becausethey'll talk about the results at office hours the next week.[5]A small improvement in navigational ability can make you a lotfaster, because it has a double effect: the path is shorter, andyou can travel faster along it when you're more certain it's theright one. That's where a lot of YC's value lies, in helping foundersget an extra increment of focus that lets them move faster. Andsince moving fast is the essence of a startup, YC in effect makesstartups more startup-like.Speed defines startups. Focus enables speed. YC improves focus.Why are founders uncertain about what to do? Partly because startupsalmost by definition are doing something new, which means no oneknows how to do it yet, or in most cases even what "it" is. Partlybecause startups are so counterintuitive generally. And partlybecause many founders, especially young and ambitious ones, havebeen trained to win the wrong way. That took me years to figureout. The educational system in most countries trains you to win byhacking the test instead of actually doing whatever it's supposedto measure. But that stops working when you start a startup. Sopart of what YC does is to retrain founders to stop trying to hackthe test. (It takes a surprisingly long time. A year in, you stillsee them reverting to their old habits.)YC is not simply more experienced founders passing on their knowledge.It's more like specialization than apprenticeship. The knowledgeof the YC partners and the founders have different shapes: Itwouldn't be worthwhile for a founder to acquire the encyclopedicknowledge of startup problems that a YC partner has, just as itwouldn't be worthwhile for a YC partner to acquire the depth ofdomain knowledge that a founder has. That's why it can still bevaluable for an experienced founder to do YC, just as it can stillbe valuable for an experienced athlete to have a coach.The other big thing YC gives founders is colleagues, and this maybe even more important than the advice of partners. If you look athistory, great work clusters around certain places and institutions:Florence in the late 15th century, the University of Göttingen inthe late 19th, The New Yorker under Ross, Bell Labs, Xerox PARC.However good you are, good colleagues make you better. Indeed, veryambitious people probably need colleagues more than anyone else,because they're so starved for them in everyday life.Whether or not YC manages one day to be listed alongside thosefamous clusters, it won't be for lack of trying. We were very awareof this historical phenomenon and deliberately designed YC to beone. By this point it's not bragging to say that it's the biggestcluster of great startup founders. Even people trying to attack YCconcede that.Colleagues and startup founders are two of the most powerful forcesin the world, so you'd expect it to have a big effect to combinethem. Before YC, to the extent people thought about the questionat all, most assumed they couldn't be combined — that lonelinesswas the price of independence. That was how it felt to us when westarted our own startup in Boston in the 1990s. We had a handfulof older people we could go to for advice (of varying quality), butno peers. There was no one we could commiserate with about themisbehavior of investors, or speculate with about the future oftechnology. I often tell founders to make something they themselveswant, and YC is certainly that: it was designed to be exactly whatwe wanted when we were starting a startup.One thing we wanted was to be able to get seed funding withouthaving to make the rounds of random rich people. That has become acommodity now, at least in the US. But great colleagues can neverbecome a commodity, because the fact that they cluster in someplaces means they're proportionally absent from the rest.Something magical happens where they do cluster though. The energyin the room at a YC dinner is like nothing else I've experienced.We would have been happy just to have one or two other startups totalk to. When you have a whole roomful it's another thing entirely.YC founders aren't just inspired by one another. They also help oneanother. That's the happiest thing I've learned about startupfounders: how generous they can be in helping one another. We noticedthis in the first batch and consciously designed YC to magnify it.The result is something far more intense than, say, a university.Between the partners, the alumni, and their batchmates, foundersare surrounded by people who want to help them, and can.Notes[1] This is why I've never liked it when people refer to YC as a"bootcamp." It's intense like a bootcamp, but the opposite instructure. Instead of everyone doing the same thing, they're eachtalking to YC partners to figure out what their specific startupneeds.[2] When I say the summer 2012 batch was broken, I mean it feltto the partners that something was wrong. Things weren't yet sobroken that the startups had a worse experience. In fact that batchdid unusually well.[3] This situation reminds me of the research showing that peopleare much better at answering questions than they are at judging howaccurate their answers are. The two phenomena feel very similar.[4] The Airbnbs were particularly good at listening — partlybecause they were flexible and disciplined, but also because they'dhad such a rough time during the preceding year. They were readyto listen.[5] The optimal unit of decisiveness depends on how long it takesto get results, and that depends on the type of problem you'resolving. When you're negotiating with investors, it could be acouple days, whereas if you're building hardware it could be months.Thanks to Trevor Blackwell, Jessica Livingston, Harj Taggar, and Garry Tan for reading drafts of this.
