Frighteningly Ambitious Startup Ideas

http://paulgraham.com/ambitious.html

March 2012

One of the more surprising things I've noticed while workingon Y Combinator is how frightening the most ambitious startupideas are. In this essay I'm going to demonstratethis phenomenon by describing some. Any one of themcould make you a billionaire. That might sound like an attractiveprospect, and yet when I describe these ideas you maynotice you find yourself shrinking away from them.

Don't worry, it's not a sign of weakness. Arguably it's a sign ofsanity. The biggest startup ideas are terrifying. And not justbecause they'd be a lot of work. The biggest ideas seem to threatenyour identity: you wonder if you'd have enough ambition to carrythem through.

There's a scene in Being John Malkovich where the nerdy heroencounters a very attractive, sophisticated woman. She says tohim:

Here's the thing: If you ever got me, you wouldn't have a clue what to do with me.

That's what these ideas say to us.

This phenomenon is one of the most important things you can understandabout startups. [1]You'd expect big startup ideas to beattractive, but actually they tend to repel you. And that has abunch of consequences. It means these ideas are invisible to mostpeople who try to think of startup ideas, because their subconsciousfilters them out. Even the most ambitious people are probably bestoff approaching them obliquely.

1. A New Search Engine

The best ideas are just on the right side of impossible. I don'tknow if this one is possible, but there are signs it might be.Making a new search engine means competing with Google, and recentlyI've noticed some cracks in their fortress.

The point when it became clear to me that Microsoft had lost theirway was when they decided to get into the search business. Thatwas not a natural move for Microsoft. They did it because theywere afraid of Google, and Google was in the search business. Butthis meant (a) Google was now setting Microsoft's agenda, and (b)Microsoft's agenda consisted of stuff they weren't good at.

Microsoft : Google :: Google : Facebook.

That does not by itself meanthere's room for a new search engine, but lately when using Googlesearch I've found myself nostalgic for the old days, whenGoogle was true to its own slightly aspy self. Google used to giveme a page of the right answers, fast, with no clutter. Now theresults seem inspired by the Scientologist principle that what'strue is what's true for you. And the pages don't have theclean, sparse feel they used to. Google search results used tolook like the output of a Unix utility. Now if I accidentally putthe cursor in the wrong place, anything might happen.

The way to win here is to build the search engine all the hackersuse. A search engine whose users consisted of the top 10,000 hackersand no one else would be in a very powerful position despite itssmall size, just as Google was when it was that search engine. Andfor the first time in over a decade the idea of switching seemsthinkable to me.

Since anyone capable of starting this company is one of those 10,000hackers, the route is at least straightforward: make the searchengine you yourself want. Feel free to make it excessively hackerish.Make it really good for code search, for example. Would you likesearch queries to be Turing complete? Anything that gets you those10,000 users is ipso facto good.

Don't worry if something you want to do will constrain you in thelong term, because if you don't get that initial core of users,there won't be a long term. If you can just build something thatyou and your friends genuinely prefer to Google, you're alreadyabout 10% of the way to an IPO, just as Facebook was (though theyprobably didn't realize it) when they got all the Harvard undergrads.

2. Replace Email

Email was not designed to be used the way we use it now. Email isnot a messaging protocol. It's a todo list. Or rather, my inboxis a todo list, and email is the way things get onto it. But itis a disastrously bad todo list.

I'm open to different types of solutions to this problem, but Isuspect that tweaking the inbox is not enough, and that email hasto be replaced with a new protocol. This new protocol should be a todo list protocol, nota messaging protocol, although there is a degenerate case wherewhat someone wants you to do is: read the following text.

As a todo list protocol, the new protocol should give more powerto the recipient than email does. I want there to be more restrictionson what someone can put on my todo list. And when someone can putsomething on my todo list, I want them to tell me more about whatthey want from me. Do they want me to do something beyond justreading some text? How important is it? (There obviously has tobe some mechanism to prevent people from saying everything isimportant.) When does it have to be done?

This is one of those ideas that's like an irresistible force meetingan immovable object. On one hand, entrenched protocols are impossibleto replace. On the other, it seems unlikely that people in100 years will still be living in the same email hell we do now.And if email is going to get replaced eventually, why not now?

If you do it right, you may be able to avoid the usual chicken and egg problem new protocols face, because some of the most powerfulpeople in the world will be among the first to switch to it. They're all at the mercy of email too.

Whatever you build, make it fast. GMail has become painfully slow.[2]If you made something no better than GMail, but fast, thatalone would let you start to pull users away from GMail.

GMail is slow because Google can't afford to spend a lot on it.But people will pay for this. I'd have no problem paying $50 a month.Considering how much time I spend in email, it's kind of scary tothink how much I'd be justified in paying. At least $1000 a month.If I spend several hours a day reading and writing email, that wouldbe a cheap way to make my life better.

3. Replace Universities

People are all over this idea lately, and I think they're ontosomething. I'm reluctant to suggest that an institution that'sbeen around for a millennium is finished just because of some mistakesthey made in the last few decades, but certainly in the last fewdecades US universities seem to have been headed down the wrongpath. One could do a lot better for a lot less money.

I don't think universities will disappear. They won't be replacedwholesale. They'll just lose the de facto monopoly on certain typesof learning that they once had. There will be many different waysto learn different things, and some may look quite different fromuniversities. Y Combinator itself is arguably one of them.

Learning is such a big problem that changing the way people do itwill have a wave of secondary effects. For example, the name ofthe university one went to is treated by a lot of people (correctlyor not) as a credential in its own right. If learning breaks upinto many little pieces, credentialling may separate from it. Theremay even need to be replacements for campus social life (and oddlyenough, YC even has aspects of that).

You could replace high schools too, but there you face bureaucraticobstacles that would slow down a startup. Universities seem theplace to start.

4. Internet Drama

Hollywood has been slow to embrace the Internet. That was a mistake, because I think we can now call a winner in the race betweendelivery mechanisms, and it is the Internet, not cable.

A lot of the reason is the horribleness of cable clients, also knownas TVs. Our family didn't wait for Apple TV. We hated our lastTV so much that a few months ago we replaced it with an iMac boltedto the wall. It's a little inconvenient to control it with awireless mouse, but the overall experience is much better than thenightmare UI we had to deal with before.

Some of the attention people currently devote to watchingmovies and TV can be stolen by things that seem completely unrelated,like social networking apps. More can be stolen by things that area little more closely related, like games. But there will probablyalways remain some residual demand for conventional drama, whereyou sit passively and watch as a plot happens. So how do you deliverdrama via the Internet? Whatever you make will have to be on alarger scale than Youtube clips. When people sit down to watch ashow, they want to know what they're going to get: either partof a series with familiar characters, or a single longer "movie"whose basic premise they know in advance.

There are two ways delivery and payment could play out. Eithersome company like Netflix or Apple will be the app store forentertainment, and you'll reach audiences through them. Or thewould-be app stores will be too overreaching, or too technicallyinflexible, and companies will arise to supply payment and streaminga la carte to the producers of drama. If that's the way thingsplay out, there will also be a need for such infrastructure companies.

5. The Next Steve Jobs

I was talking recently to someone who knew Apple well, and I askedhim if the people now running the company would be able to keepcreating new things the way Apple had under Steve Jobs. His answerwas simply "no." I already feared that would be the answer. Iasked more to see how he'd qualify it. But he didn't qualify itat all. No, there will be no more great new stuff beyond whatever'scurrently in the pipeline. Apple'srevenues may continue to rise for a long time, but as Microsoftshows, revenue is a lagging indicator in the technology business.

So if Apple's not going to make the next iPad, who is? None of theexisting players. None of them are run by product visionaries, andempirically you can't seem to get those by hiring them. Empiricallythe way you get a product visionary as CEO is for him to found thecompany and not get fired. So the company that creates the nextwave of hardware is probably going to have to be a startup.

I realize it sounds preposterously ambitious for a startup to tryto become as big as Apple. But no more ambitious than it was forApple to become as big as Apple, and they did it. Plus a startuptaking on this problem now has an advantage the original Appledidn't: the example of Apple. Steve Jobs has shown us what'spossible. That helps would-be successors both directly, as RogerBannister did, by showing how much better you can do than peopledid before, and indirectly, as Augustus did, by lodging the ideain users' minds that a single person could unroll the future for them. [3]

Now Steve is gone there's a vacuum we can all feel. If a new companyled boldly into the future of hardware, users would follow. TheCEO of that company, the "next Steve Jobs," might not measure upto Steve Jobs. But he wouldn't have to. He'd just have to do abetter job than Samsung and HP and Nokia, and that seems prettydoable.

6. Bring Back Moore's Law

The last 10 years have reminded us what Moore's Law actually says.Till about 2002 you could safely misinterpret it as promising thatclock speeds would double every 18 months. Actually what it saysis that circuit densities will double every 18 months. It used toseem pedantic to point that out. Not any more. Intel can no longergive us faster CPUs, just more of them.

This Moore's Law is not as good as the old one. Moore's Law usedto mean that if your software was slow, all you had to do was wait,and the inexorable progress of hardware would solve your problems.Now if your software is slow you have to rewrite it to do morethings in parallel, which is a lot more work than waiting.

It would be great if a startup could give us something of the oldMoore's Law back, by writing software that could make a large numberof CPUs look to the developer like one very fast CPU. There areseveral ways to approach this problem. The most ambitious is totry to do it automatically: to write a compiler that will parallelizeour code for us. There's a name for this compiler, the sufficientlysmart compiler, and it is a byword for impossibility. But isit really impossible? Is there no configuration of the bits inmemory of a present day computer that is this compiler? If youreally think so, you should try to prove it, because that would bean interesting result. And if it's not impossible but simply veryhard, it might be worth trying to write it. The expected valuewould be high even if the chance of succeeding was low.

The reason the expected value is so high is web services. If youcould write software that gave programmers the convenience of theway things were in the old days, you could offer it to them as aweb service. And that would in turn mean that you got practicallyall the users.

Imagine there was another processor manufacturer that could still translateincreased circuit densities into increased clock speeds. They'dtake most of Intel's business. And since web services mean thatno one sees their processors anymore, by writing the sufficientlysmart compiler you could create a situation indistinguishable fromyou being that manufacturer, at least for the server market.

The least ambitious way of approaching the problem is to start fromthe other end, and offer programmers more parallelizable Lego blocksto build programs out of, like Hadoop and MapReduce. Then theprogrammer still does much of the work of optimization.

There's an intriguing middle ground where you build a semi-automaticweapon—where there's a human in the loop. You make somethingthat looks to the user like the sufficiently smart compiler, butinside has people, using highly developed optimization tools tofind and eliminate bottlenecks in users' programs. These peoplemight be your employees, or you might create a marketplace foroptimization.

An optimization marketplace would be a way to generate the sufficientlysmart compiler piecemeal, because participants would immediatelystart writing bots. It would be a curious state of affairs if youcould get to the point where everything could be done by bots,because then you'd have made the sufficiently smart compiler, butno one person would have a complete copy of it.

I realize how crazy all this sounds. In fact, what I like aboutthis idea is all the different ways in which it's wrong. The wholeidea of focusing on optimization is counter to the general trendin software development for the last several decades. Trying towrite the sufficiently smart compiler is by definition a mistake.And even if it weren't, compilers are the sort of software that'ssupposed to be created by open source projects, not companies. Plusif this works it will deprive all the programmers who take pleasurein making multithreaded apps of so much amusing complexity. Theforum troll I have by now internalized doesn't even know where tobegin in raising objections to this project. Now that's what Icall a startup idea.

7. Ongoing Diagnosis

But wait, here's another that could face even greater resistance:ongoing, automatic medical diagnosis.

One of my tricks for generating startup ideas is to imagine theways in which we'll seem backward to future generations. And I'mpretty sure that to people 50 or 100 years in the future, it willseem barbaric that people in our era waited till they had symptomsto be diagnosed with conditions like heart disease and cancer.

For example, in 2004 Bill Clinton found he was feeling short ofbreath. Doctors discovered that several of his arteries were over90% blocked and 3 days later he had a quadruple bypass. It seemsreasonable to assume Bill Clinton has the best medical care available.And yet even he had to wait till his arteries were over 90% blockedto learn that the number was over 90%. Surely at some point in thefuture we'll know these numbers the way we now know something likeour weight. Ditto for cancer. It will seem preposterous to futuregenerations that we wait till patients have physical symptoms tobe diagnosed with cancer. Cancer will show up on some sort of radarscreen immediately.

(Of course, what shows up on the radar screen may be different fromwhat we think of now as cancer. I wouldn't be surprised if at anygiven time we have ten or even hundreds of microcancers going atonce, none of which normally amount to anything.)

A lot of the obstacles to ongoing diagnosis will come from the factthat it's going against the grain of the medical profession. Theway medicine has always worked is that patients come to doctorswith problems, and the doctors figure out what's wrong. A lot ofdoctors don't like the idea of going on the medical equivalent ofwhat lawyers call a "fishing expedition," where you go looking forproblems without knowing what you're looking for. They call thethings that get discovered this way "incidentalomas," and they aresomething of a nuisance.

For example, a friend of mine once had her brain scanned as partof a study. She was horrified when the doctors running the studydiscovered what appeared to be a large tumor. After further testing,it turned out to be a harmless cyst. But it cost her a few daysof terror. A lot of doctors worry that if you start scanning peoplewith no symptoms, you'll get this on a giant scale: a huge numberof false alarms that make patients panic and require expensive andperhaps even dangerous tests to resolve. But I think that's justan artifact of current limitations. If people were scanned all thetime and we got better at deciding what was a real problem, myfriend would have known about this cyst her whole life and knownit was harmless, just as we do a birthmark.

There is room for a lot of startups here. In addition to the technical obstacles allstartups face, and the bureaucratic obstacles all medical startupsface, they'll be going against thousands of years of medicaltradition. But it will happen, and it will be a great thing—sogreat that people in the future will feel as sorry for us as we dofor the generations that lived before anaesthesia and antibiotics.

Tactics

Let me conclude with some tactical advice. If you want to take ona problem as big as the ones I've discussed, don't make a directfrontal attack on it. Don't say, for example, that you're goingto replace email. If you do that you raise too many expectations.Your employees and investors will constantly be asking "are we thereyet?" and you'll have an army of haters waiting to see you fail.Just say you're building todo-list software. That sounds harmless.People can notice you've replaced email when it's a fait accompli.[4]

Empirically, the way to do really big things seems to be to startwith deceptively small things. Want to dominate microcomputersoftware? Start by writing a Basic interpreter for a machine witha few thousand users. Want to make the universal web site? Startby building a site for Harvard undergrads to stalk one another.

Empirically, it's not just for other people that you need to startsmall. You need to for your own sake. Neither Bill Gates nor MarkZuckerberg knew at first how big their companies were going to get.All they knew was that they were onto something. Maybe it's a badidea to have really big ambitions initially, because the biggeryour ambition, the longer it's going to take, and the further youproject into the future, the more likely you'll get it wrong.

I think the way to use these big ideas is not to try to identify aprecise point in the future and then ask yourself how to get fromhere to there, like the popular image of a visionary. You'll bebetter off if you operate like Columbus and just head in a generalwesterly direction. Don't try to construct the future like abuilding, because your current blueprint is almost certainly mistaken.Start with something you know works, and when you expand, expandwestward.

The popular image of the visionary is someone with a clear view ofthe future, but empirically it may be better to have a blurry one.





Notes

[1]It's also one of the most important things VCs fail tounderstand about startups. Most expect founders to walk in with aclear plan for the future, and judge them based on that. Fewconsciously realize that in the biggest successes there is the leastcorrelation between the initial plan and what the startup eventuallybecomes.

[2]This sentence originally read "GMail is painfully slow."Thanks to Paul Buchheit for the correction.

[3]Roger Bannister is famous as the first person to run a milein under 4 minutes. But his world record only lasted 46 days. Oncehe showed it could be done, lots of others followed. Ten yearslater Jim Ryun ran a 3:59 mile as a high school junior.

[4]If you want to be the next Apple, maybe you don't even want to startwith consumer electronics. Maybe at first you make something hackersuse. Or you make something popular but apparently unimportant,like a headset or router. All you need is a bridgehead.

Thanks to Sam Altman, Trevor Blackwell, Paul Buchheit, Patrick Collison, Aaron Iba, JessicaLivingston, Robert Morris, Harj Taggar and Garry Tanfor reading drafts of this.