October 2005(This essay is derived from a talk at the 2005 Startup School.)How do you get good ideas for startups? That's probably the numberone question people ask me.I'd like to reply with another question: why do people think it'shard to come up with ideas for startups?That might seem a stupid thing to ask. Why do they thinkit's hard? If people can't do it, then it is hard, at leastfor them. Right?Well, maybe not. What people usually say is not that they can'tthink of ideas, but that they don't have any. That's not quite thesame thing. It could be the reason they don't have any is thatthey haven't tried to generate them.I think this is often the case. I think people believe that comingup with ideas for startups is very hard-- that it must bevery hard-- and so they don't try do to it. They assume ideas arelike miracles: they either pop into your head or they don't.I also have a theory about why people think this. They overvalueideas. They think creating a startup is just a matter of implementingsome fabulous initial idea. And since a successful startup is worthmillions of dollars, a good idea is therefore a million dollar idea.If coming up with an idea for a startup equals coming up with amillion dollar idea, then of course it's going to seem hard. Toohard to bother trying. Our instincts tell us something so valuablewould not be just lying around for anyone to discover.Actually, startup ideas are not million dollar ideas, and here'san experiment you can try to prove it: just try to sell one. Nothingevolves faster than markets. The fact that there's no market forstartup ideas suggests there's no demand. Which means, in thenarrow sense of the word, that startup ideas are worthless.QuestionsThe fact is, most startups end up nothing like the initial idea.It would be closer to the truth to say the main value of your initialidea is that, in the process of discovering it's broken, you'llcome up with your real idea.The initial idea is just a starting point-- not a blueprint, but aquestion. It might help if they were expressed that way. Insteadof saying that your idea is to make a collaborative, web-basedspreadsheet, say: could one make a collaborative, web-basedspreadsheet? A few grammatical tweaks, and a woefully incompleteidea becomes a promising question to explore.There's a real difference, because an assertion provokes objectionsin a way a question doesn't. If you say: I'm going to build aweb-based spreadsheet, then critics-- the most dangerous of whichare in your own head-- will immediately reply that you'd be competingwith Microsoft, that you couldn't give people the kind of UI theyexpect, that users wouldn't want to have their data on your servers,and so on.A question doesn't seem so challenging. It becomes: let's trymaking a web-based spreadsheet and see how far we get. And everyoneknows that if you tried this you'd be able to make somethinguseful. Maybe what you'd end up with wouldn't even be a spreadsheet.Maybe it would be some kind of new spreasheet-like collaborationtool that doesn't even have a name yet. You wouldn't have thoughtof something like that except by implementing your way toward it.Treating a startup idea as a question changes what you're lookingfor. If an idea is a blueprint, it has to be right. But if it'sa question, it can be wrong, so long as it's wrong in a way thatleads to more ideas.One valuable way for an idea to be wrong is to be only a partialsolution. When someone's working on a problem that seems toobig, I always ask: is there some way to bite off some subset of theproblem, then gradually expand from there? That will generallywork unless you get trapped on a local maximum, like 1980s-styleAI, or C.UpwindSo far, we've reduced the problem from thinking of a million dollaridea to thinking of a mistaken question. That doesn't seem so hard,does it?To generate such questions you need two things: to be familiar with promising new technologies, and to have the right kind of friends.New technologies are the ingredients startup ideas are made of, andconversations with friends are the kitchen they're cooked in.Universities have both, and that's why so many startups grow outof them. They're filled with new technologies, because they'retrying to produce research, and only things that are new count asresearch. And they're full of exactly the right kind of people to have ideas with: the other students, who will be not only smart butelastic-minded to a fault.The opposite extreme would be a well-paying but boring job at a bigcompany. Big companies are biased against new technologies, andthe people you'd meet there would be wrong too.In an essay I wrote for high school students, I said a good rule of thumb was to stay upwind-- towork on things that maximize your future options. The principleapplies for adults too, though perhaps it has to be modified to:stay upwind for as long as you can, then cash in the potentialenergy you've accumulated when you need to pay for kids.I don't think people consciously realize this, but one reasondownwind jobs like churning out Java for a bank pay so well is precisely that they are downwind. The market price for that kindof work is higher because it gives you fewer options for the future.A job that lets you work on exciting new stuff will tend to payless, because part of the compensation is in the form of the newskills you'll learn.Grad school is the other end of the spectrum from a coding job ata big company: the pay's low but you spend most of your time workingon new stuff. And of course, it's called "school," which makesthat clear to everyone, though in fact all jobs are some percentageschool.The right environment for having startup ideas need not be auniversity per se. It just has to be a situation with a largepercentage of school.It's obvious why you want exposure to new technology, but why do you need other people? Can't you just think of new ideas yourself?The empirical answer is: no. Even Einstein needed people to bounceideas off. Ideas get developed in the process of explaining themto the right kind of person. You need that resistance, justas a carver needs the resistance of the wood.This is one reason Y Combinator has a rule against investing in startups with only one founder. Practically every successful companyhas at least two. And because startup founders work under great pressure, it's critical they be friends.I didn't realize it till I was writing this, but that may helpexplain why there are so few female startup founders. I read onthe Internet (so it must be true) that only 1.7% of VC-backedstartups are founded by women. The percentage of female hackersis small, but not that small. So why the discrepancy?When you realize that successful startups tend to have multiplefounders who were already friends, apossible explanation emerges. People's best friends are likely to be of the same sex, and if one group is a minority in some population,pairs of them will be a minority squared.[1]DoodlingWhat these groups of co-founders do together is more complicated than just sitting down and trying to think of ideas. I suspect the most productive setup is a kind of together-alone-together sandwich.Together you talk about some hard problem, probably getting nowhere.Then, the next morning, one of you has an idea in the shower abouthow to solve it. He runs eagerly to to tell the others, and togetherthey work out the kinks.What happens in that shower? It seems to me that ideas just popinto my head. But can we say more than that?Taking a shower is like a form of meditation. You're alert, butthere's nothing to distract you. It's in a situation like this,where your mind is free to roam, that it bumps into new ideas.What happens when your mind wanders? It may be like doodling. Mostpeople have characteristic ways of doodling. This habit is unconscious, but not random: I found my doodles changed after I started studying painting. I started to make the kind of gesturesI'd make if I were drawing from life. They were atoms of drawing, but arranged randomly.[2]Perhaps letting your mind wander is like doodling with ideas. Youhave certain mental gestures you've learned in your work, and whenyou're not paying attention, you keep making these same gestures, but somewhat randomly. In effect, you call the same functions onrandom arguments. That's what a metaphor is: a function applied to an argument of the wrong type.Conveniently, as I was writing this, my mind wandered: would it beuseful to have metaphors in a programming language? I don't know;I don't have time to think about this. But it's convenient becausethis is an example of what I mean by habits of mind. I spend a lotof time thinking about language design, and my habit of always asking "would x be useful in a programming language" just gotinvoked.If new ideas arise like doodles, this would explain why you haveto work at something for a while before you have any. It's notjust that you can't judge ideas till you're an expert in a field.You won't even generate ideas, because you won't have any habitsof mind to invoke.Of course the habits of mind you invoke on some field don't haveto be derived from working in that field. In fact, it's oftenbetter if they're not. You're not just looking for good ideas, butfor good new ideas, and you have a better chance of generatingthose if you combine stuff from distant fields. As hackers, oneof our habits of mind is to ask, could one open-source x? For example, what if you made an open-source operating system? A fineidea, but not very novel. Whereas if you ask, could you make anopen-source play? you might be onto something.Are some kinds of work better sources of habits of mind than others?I suspect harder fields may be better sources, because to attackhard problems you need powerful solvents. I find math is a goodsource of metaphors-- good enough that it's worth studying just forthat. Related fields are also good sources, especially when they'rerelated in unexpected ways. Everyone knows computer science andelectrical engineering are related, but precisely because everyoneknows it, importing ideas from one to the other doesn't yield greatprofits. It's like importing something from Wisconsin to Michigan. Whereas (I claim) hacking and painting arealso related, in the sense that hackers and painters are both makers,and this source of new ideas is practically virgin territory.ProblemsIn theory you could stick together ideas at random and see what youcame up with. What if you built a peer-to-peer dating site? Wouldit be useful to have an automatic book? Could you turn theoremsinto a commodity? When you assemble ideas at random like this, they may not be just stupid, but semantically ill-formed. What would it even mean to make theorems a commodity? You got me. Ididn't think of that idea, just its name.You might come up with something useful this way, but I never have.It's like knowing a fabulous sculpture is hidden inside a block ofmarble, and all you have to do is remove the marble that isn't partof it. It's an encouraging thought, because it reminds you there is an answer, but it's not much use in practice because the searchspace is too big.I find that to have good ideas I need to be working on some problem.You can't start with randomness. You have to start with a problem,then let your mind wander just far enough for new ideas to form.In a way, it's harder to see problems than their solutions. Most people prefer to remain in denial about problems. It's obviouswhy: problems are irritating. They're problems! Imagine if peoplein 1700 saw their lives the way we'd see them. It would have beenunbearable. This denial is such a powerful force that, even when presented with possible solutions, people often prefer to believethey wouldn't work.I saw this phenomenon when I worked on spam filters. In 2002, mostpeople preferred to ignore spam, and most of those who didn'tpreferred to believe the heuristic filters then available were thebest you could do.I found spam intolerable, and I felt it had to be possible torecognize it statistically. And it turns out that was all you needed to solve the problem. The algorithm I used was ridiculouslysimple. Anyone who'd really tried to solve the problem would havefound it. It was just that no one had really tried to solve theproblem.[3]Let me repeat that recipe: finding the problem intolerable and feeling it must be possible to solve it. Simple as it seems, that'sthe recipe for a lot of startup ideas.WealthSo far most of what I've said applies to ideas in general. What's special about startup ideas? Startup ideas are ideas for companies,and companies have to make money. And the way to make money is tomake something people want.Wealth is what people want. I don't mean that as some kind of philosophical statement; I mean it as a tautology.So an idea for a startup is an idea for something people want.Wouldn't any good idea be something people want? Unfortunately not. I think new theorems are a fine thing to create, but thereis no great demand for them. Whereas there appears to be greatdemand for celebrity gossip magazines. Wealth is defined democratically.Good ideas and valuable ideas are not quite the same thing; thedifference is individual tastes.But valuable ideas are very close to good ideas, especially intechnology. I think they're so close that you can get away withworking as if the goal were to discover good ideas, so long as, inthe final stage, you stop and ask: will people actually pay for this? Only a few ideas are likely to make it that far and then getshot down; RPN calculators might be one example.One way to make something people want is to look at stuff people use now that's broken. Dating sites are a prime example. They have millions of users, so they must be promising something people want. And yet they work horribly. Just ask anyone who uses them. It's as if they used the worse-is-better approach but stopped afterthe first stage and handed the thing over to marketers.Of course, the most obvious breakage in the average computer user's life is Windows itself. But this is a special case: you can'tdefeat a monopoly by a frontal attack. Windows can and will be overthrown, but not by giving people a better desktop OS. The wayto kill it is to redefine the problem as a superset of the current one. The problem is not, what operating system should people useon desktop computers? but how should people use applications?There are answers to that question that don't even involve desktopcomputers.Everyone thinks Google is going to solve this problem, but it is avery subtle one, so subtle that a company as big as Google mightwell get it wrong. I think the odds are better than 50-50 that theWindows killer-- or more accurately, Windows transcender-- willcome from some little startup.Another classic way to make something people want is to take aluxury and make it into a commmodity. People must want somethingif they pay a lot for it. And it is a very rare product that can'tbe made dramatically cheaper if you try.This was Henry Ford's plan. He made cars, which had been a luxuryitem, into a commodity. But the idea is much older than Henry Ford.Water mills transformed mechanical power from a luxury into acommodity, and they were used in the Roman empire. Arguablypastoralism transformed a luxury into a commodity.When you make something cheaper you can sell more of them. But ifyou make something dramatically cheaper you often get qualitativechanges, because people start to use it in different ways. Forexample, once computers get so cheap that most people can have oneof their own, you can use them as communication devices.Often to make something dramatically cheaper you have to redefine the problem. The Model T didn't have all the features previouscars did. It only came in black, for example. But it solved theproblem people cared most about, which was getting from place toplace.One of the most useful mental habits I know I learned from MichaelRabin: that the best way to solve a problem is often to redefineit. A lot of people use this technique without being consciouslyaware of it, but Rabin was spectacularly explicit. You need a bigprime number? Those are pretty expensive. How about if I give youa big number that only has a 10 to the minus 100 chance of not beingprime? Would that do? Well, probably; I mean, that's probablysmaller than the chance that I'm imagining all this anyway.Redefining the problem is a particularly juicy heuristic when youhave competitors, because it's so hard for rigid-minded people to follow. You can work in plain sight and they don't realize the danger. Don't worry about us. We're just working on search. Do one thing and do it well, that's our motto.Making things cheaper is actually a subset of a more generaltechnique: making things easier. For a long time it was most of making things easier, but now that the things we build are socomplicated, there's another rapidly growing subset: making things easier to use.This is an area where there's great room for improvement. What youwant to be able to say about technology is: it just works. Howoften do you say that now?Simplicity takes effort-- genius, even. The average programmer seems to produce UI designs that are almost willfully bad. I was trying to use the stove at my mother's house a couple weeks ago. It was a new one, and instead of physical knobs it had buttons andan LED display. I tried pressing some buttons I thought would causeit to get hot, and you know what it said? "Err." Not even "Error.""Err." You can't just say "Err" to the user of a stove.You should design the UI so that errors are impossible. And the boneheads who designed this stove even had an example of such a UIto work from: the old one. You turn one knob to set the temperatureand another to set the timer. What was wrong with that? It justworked.It seems that, for the average engineer, more options just meansmore rope to hang yourself. So if you want to start a startup, youcan take almost any existing technology produced by a big company, and assume you could build something way easier to use.Design for ExitSuccess for a startup approximately equals getting bought. Youneed some kind of exit strategy, because you can't get the smartestpeople to work for you without giving them options likely to beworth something. Which means you either have to get bought or gopublic, and the number of startups that go public is very small.If success probably means getting bought, should you make that aconscious goal? The old answer was no: you were supposed to pretendthat you wanted to create a giant, public company, and act surprisedwhen someone made you an offer. Really, you want to buy us? Well,I suppose we'd consider it, for the right price.I think things are changing. If 98% of the time success means getting bought, why not be open about it? If 98% of the time you'redoing product development on spec for some big company, why notthink of that as your task? One advantage of this approach is thatit gives you another source of ideas: look at big companies, thinkwhat they should be doing, and do it yourself. Even ifthey already know it, you'll probably be done faster.Just be sure to make something multiple acquirers will want. Don'tfix Windows, because the only potential acquirer is Microsoft, and when there's only one acquirer, they don't have to hurry. They cantake their time and copy you instead of buying you. If you wantto get market price, work on something where there's competition.If an increasing number of startups are created to do productdevelopment on spec, it will be a natural counterweight to monopolies.Once some type of technology is captured by a monopoly, it will only evolve at big company rates instead of startup rates, whereasalternatives will evolve with especial speed. A free marketinterprets monopoly as damage and routes around it.The Woz RouteThe most productive way to generate startup ideas is also themost unlikely-sounding: by accident. If you look at how famousstartups got started, a lot of them weren't initially supposed to be startups. Lotus began with a program Mitch Kapor wrote for afriend. Apple got started because Steve Wozniak wanted to buildmicrocomputers, and his employer, Hewlett-Packard, wouldn't let himdo it at work. Yahoo began as David Filo's personal collection oflinks.This is not the only way to start startups. You can sit down andconsciously come up with an idea for a company; we did. But measuredin total market cap, the build-stuff-for-yourself model might be more fruitful. It certainly has to be the most fun way to come upwith startup ideas. And since a startup ought to have multiplefounders who were already friends before they decided to start a company, the rather surprising conclusion is that the best way to generate startup ideas is to do what hackers do for fun: cook upamusing hacks with your friends.It seems like it violates some kind of conservation law, but thereit is: the best way to get a "million dollar idea" is just to dowhat hackers enjoy doing anyway.Notes[1]This phenomenon may account for a number of discrepanciescurrently blamed on various forbidden isms. Never attribute to malice what can be explained by math.[2] A lot of classic abstract expressionism is doodling of this type:artists trained to paint from life using the same gestures butwithout using them to represent anything. This explains why suchpaintings are (slightly) more interesting than random marks would be.[3]Bill Yerazunis had solved the problem, but he got there byanother path. He made a general-purpose file classifier so goodthat it also worked for spam.
