defmacro

nobody knows the future

2020-06-02T00:00:00+00:00

Can you recall the experience of waking up for the first time in a new city? Your senses are filled with sights and sounds that you couldn’t easily relay to someone, not without a lot of poetic talent, yet they’re as real to you as a block of wood? You know how you never quite forget those sights and sounds? If you wake up in this city again, even twenty years later, your senses awaken all those memories you didn’t think you had in you anymore.

I was born in a country that does not exist today. I was only a child, but I still remember the world of doublespeak and tyranny, the world of budding freedom, and the world of tragedy and chaos. Like cities, each of these worlds has a sound and a smell. I couldn’t relay them to you here, but they’re very real to me. As real as anything I’ve ever experienced. The smells and sounds that are in the air now are familiar. All the memories I didn’t think I’d remembered are rushing in.

What I can relay to you is that some of these sounds, these smells, these memories are not good. Not good for you, not good for me, not good for the disenfranchised people, not good for George Floyd’s family, not good for the world. The intentions are good, but good intentions aren’t enough– history is littered with countless examples when good intentions led to unthinkable tragedy.

From here, the road we’re collectively on forks into many different paths. Some of them lead to tragedy of unimaginable scale, and some lead to a better world for all of us. Everyone around us– the democrats, the republicans, the protest organizers, celebrities, our political leaders, tech CEOs, Twitter bots, friends, family members– everyone, seems incredibly certain about which path we should take. But among this sea of certainty I feel incredibly uncertain.

There is only one thing I know with absolute certainty, and it’s that nobody– not the democrats, not the republicans, not the protest organizers, not the celebrities, not our political leaders, not the tech CEOs, not Twitter bots, not friends, not family members– nobody, knows which paths lead to tragedy and which paths lead to a just, prosperous world. All of them are pulling us along with absolute zeal and boundless energy, striking iron while it’s hot, many without giving a single thought to the sparks that may ignite America and the planet.

I do not advocate for slowing down the progress of civil rights, or waiting for a more opportune moment, or for staying moderate in the face of injustice. All I ask is that we trust our leaders and our passions a little less, and trust our hearts and our neighbors a little more.

I am an atheist, but I am praying we choose peace and prosperity, not violence and annihilation. Nobody knows the future.

Radical markets

2019-11-26T02:02:02+00:00

In the short run, the market is a voting machine but in the long run, it is a weighing machine.
- Benjamin Graham

Radical markets is a technique that illuminates complex problems by parameterizing nearly anything in the world by price (e.g. bananas, scientific progress, time, human life), and letting market participants find the equilibrium between buyers and sellers. One way to think of this technique is as applied microeconomics. Prices carry information about people’s beliefs, and are difficult to consistently falsify; so radical markets is a means of solving problems by capturing aggregate information.

Illustrative example

My favorite example of radical markets comes from a Harvard statistician Xiao-Li Meng (h/t David Robinson). Meng proposed a market solution to the following problem: which p-value cutoff should scientists use for hypothesis testing? If the cutoff is too low, science will progress slower than it needs to because the standard of evidence is unnecessarily high. If the cutoff is too high, we may find ourselves in a replication crisis. This problem is further exacerbated by p-hacking– scientists are incentivized to understate the risk of false positives.

Meng proposed the following approach– you can use whatever cutoff you want, but your salary will be cut by the corresponding level every time your claim turns out to be wrong.

Your first reaction might be “this will never work!”, but the more you think about this proposal, the more its benefits become apparent. It eliminates p-hacking incentives, discourages publishing results that scientists know ex ante are unlikely to be replicated, broadly aligns incentives of individual scientists with aggregate scientific progress, and eliminates the need for magic constants. Scientists who are convinced in their results but don’t have sufficient evidence can credibly signal their confidence to the scientific community by picking a high cutoff.

More examples

Here is an example of a similar problem in a completely different domain: how should the government assess the value of property for the purpose of taxation? Via Should we move to self-assessed property taxation:

The core proposal is you announce how much each piece of your property is worth, and you are then taxed as a percentage of that value (say 2.5%). At the same time, you have to sell your property for that same value, if someone bids for it, thereby lowering or eliminating the incentive to under-report true values. If you think this through, you can see it minimizes holdout problems.

More on this at Between Property and Liability:

This would avoid administrative property valuations, discourage people from sitting on stuff they don’t use, and make it much easier to assemble property into large units. Eminent domain would no longer be needed.

(The authors of the proposal Eric Posner and Glen Weyl have a book– Radical Markets– with even more examples of this technique).

Note that you don’t necessarily need to set up a market– you can use this technique to run a thought experiment. (This isn’t as good of course, as there is a huge difference between actually choosing to pay vs hypothesizing, but running the thought experiment still gives you a lot of information.) As an example, in radical quantification we dealt with a question: how much richer is an average American in 1992 than an average American in 1800? There I quoted from William Nordhaus’s approach to measure the number of hours necessary to work to produce a lux of light.

Another approach would be to pay people to move to 1800, and use the price buyers are willing to accept as information. There is no time machine, so we can’t run the experiment. But we can run a thought experiment. Here is an example I tried on Twitter:

The methodology here is really simple, but it does give us some information– most people who chose to participate in the poll seem to value the technological advancements between 1800 and now at a very high multiple.

A more general example of this is the idea of prediction markets. The conceit is that on average a betting market will give you a better prediction of the future (and therefore a better mechanism to make decisions) than mechanisms in existing use:

Prediction markets continue to offer great potential to improve society at many levels. Their greatest promise lies in helping organizations to better aggregate info to enable better key decisions. […] such markets have consistently performed well in terms of cost, accuracy, ease of use, and user satisfaction […]

Robin Hanson also proposed Futarchy– a hypothetical system of government based on prediction markets:

In futarchy, democracy would continue to say what we want, but betting markets would now say how to get it. That is, elected representatives would formally define and manage an after-the-fact measurement of national welfare, while market speculators would say which policies they expect to raise national welfare. The basic rule of government would be: When a betting market clearly estimates that a proposed policy would increase expected national welfare, that proposal becomes law.

Another wonderful example of radical markets is by the Harvard anthropologist Joseph Henrich: Does Culture Matter in Economic Behavior? Ultimatum Game Bargaining Among the Machiguenga of the Peruvian Amazon. Henrich uses the ultimatum game to explore (and quantify) the concept of fairness and obligation across different cultures:

One player, the proposer, is endowed with a sum of money. The proposer is tasked with splitting it with another player, the responder. Once the proposer communicates their decision, the responder may accept it or reject it. If the responder accepts, the money is split per the proposal; if the responder rejects, both players receive nothing. Both players know in advance the consequences of the responder accepting or rejecting the offer.

The paper is worth reading in full, but here is a salient quote (edited for context):

Roth, in examining the difference found between American and Israeli proposers, suggests that the results indicate a difference in what is perceived as “fair,” or what is “expected” under the circumstances. My comparison of Machiguenga and Los Angeles subjects yields a similar conclusion, only more extreme. Machiguenga proposers seem to possess little or no sense of obligation to provide an equal share to responders, and responders had little or no expectation of receiving an equal share nor any desire to punish unequal divisions. The modal offer of 15 percent seemed quite “fair” to most Machiguenga.

This post is part of the Creative Library– a series of clever techniques that simplify solving complex problems.

Creative library

2019-11-26T01:01:01+00:00

Around 2016 I ran a little experiment– every time I ran into an especially clever solution to a complex problem, I saved it on my computer. In a few months I had about a dozen of these, and I noticed that the solutions repeat a small set of techniques. So I categorized each solution by technique, and gave every technique a name. At the time of this writing I have a few hundred solutions saved, categorized into 23 techniques.

Many of these techniques can be mixed and matched. When I run into a tricky problem, I go through the list of techniques and see if I can apply any to my problem. Since many of the techniques can be combined, I do this repeatedly until I can’t. More often than not this process results in an elegant solution to the problem, or at least a promising start.

When I share my approaches to these problems, people often say “that’s clever! How did you possibly think of that?” Simple! I just go through my list of techniques.

Over the next few months I’ll write up the techniques and link to them here, so this page can act as an index. Subscribe via rss or twitter to see the techniques as I post them, or just check back to this page.

Radical quantification

2019-11-26T00:00:00+00:00

When you cannot express it in numbers, your knowledge is of a meagre and unsatisfactory kind.
- Kelvin’s Dictum

Yes, and when you can express it in numbers your knowledge is of a meagre and unsatisfactory kind.
- Frank Knight

Radical quantification is a technique for dramatically simplifying complex problems with no obvious angle of attack. It works by picking just one quantity that illuminates the essence of the problem (or at least gives you a good running start). In a sense it’s a form of dimensionality reduction– like doing principal component analysis and then looking only at the first component.

Illustrative example

To this day my first encounter with this technique is the most brilliant use of it I’ve ever seen– it’s William Nordhaus’s 1996 paper Do real-output and real-wage measures capture reality? The history of lighting suggests not (Nordhaus went on to win the Nobel Prize in economics in 2018 for his work on climate change). In the paper he attempts to answer this question: how much richer is an average American in 1992 than an average American in 1800? He explains why it’s an incredibly difficult problem:

The estimates of real income are only as good as the price indexes are accurate. During periods of major technological change, the construction of accurate price indexes that capture the impact of new technologies on living standards is beyond the practical capability of official statistical agencies.

His solution? Reduce the problem to one dimension: how many hours of work did it take to produce one lux of light? Light is an especially clever metric because the technology has been with us for half a million years, has always been in demand, and has continually tracked technological advancement. Nordhaus then goes on to do incredible detective work to determine the cost of light over time; the whole paper is worth reading, but here is the conclusion:

In terms of living standards, the conventional growth in real wages has been by a factor of 13 over the 1800-1992 period. For the low-bias case, real wages have grown by a factor of 40, while in the high-bias case real wages have grown by a factor of 190.

More examples

A more recent example of radical quantification: How Superhuman Built an Engine to Find Product/Market Fit. In this post Rahul Vohra, Superhuman’s CEO, explains his methodology for measuring product/market fit, and running his whole company to maximize this metric. This use of radical quantification is incredible because conventional wisdom is that product/market fit is a qualitative endeavor. It also reminds me of How to Measure Anything: Finding the Value of Intangibles in Business.

Another example use of radical quantification to answer a qualitative question: “how important will self-driving cars be as a technology?” There is of course no way to know, but in Artificial intelligence and the modern productivity paradox: a clash of expectations and statistics Brynjolfsson et al get a handle on it by comparing the reduction in automotive jobs to the total number of jobs in America:

According to the US Bureau of Labor Statistics, in 2016 there were 3.5 million people working in private industry as “motor vehicle operators” of one sort or another (this includes truck drivers, taxi drivers, bus drivers, and other similar occupations). Suppose autonomous vehicles were to reduce, over some period, the number of drivers necessary to do the current workload to 1.5 million. We do not think this is a far-fetched scenario given the potential of the technology. Total nonfarm private employment in mid-2016 was 122 million. Therefore, autonomous vehicles would reduce the number of workers necessary to achieve the same output to 120 million. This would result in aggregate labor productivity (calculated using the standard BLS nonfarm private series) increasing by 1.7 percent (= 122/120). Supposing this transition occurred over 10 years, this single technology would provide a direct boost of 0.17 percent to annual productivity growth over that decade.

Discretizing a continuous quantity

One interesting variation of this approach is to take a continuous quantity and discretize it. Paul Graham does this in Life is Short to get a visceral sense of how short life really is:

Having kids showed me how to convert a continuous quantity, time, into discrete quantities. You only get 52 weekends with your 2 year old. If Christmas-as-magic lasts from say ages 3 to 10, you only get to watch your child experience it 8 times. And while it’s impossible to say what is a lot or a little of a continuous quantity like time, 8 is not a lot of something. If you had a handful of 8 peanuts, or a shelf of 8 books to choose from, the quantity would definitely seem limited, no matter what your lifespan was.

Tim Urban does something very similar in Your Life in Weeks. He creates a life calendar (which you can buy) where each row has 52 columns (weeks), and there are a total of 90 rows (about the average life expectancy). Looking at the calendar gives you a visceral sense of how long your life really is.

A different example of discretizing a continuous quantity is the idea of an inferential step by Eliezer Yudkowsky. Consider the question of “how much more knowledge does Alice have over Bob?” This is a vague and difficult question; to answer it you have to figure out what exactly knowledge is and how to quantity it. But an inferential step is a specific concept or an abstraction that Alice can explain to Bob. And it’s much easier to count the number of inferential steps than to measure knowledge.

This post is part of the Creative Library– a series of clever techniques that simplify solving complex problems.

The mundane, the mystical, and the meta

2019-11-25T03:03:21+00:00

In craftsman, executive I wrote about two modes of working adapted to different circumstances:

Craftsman mode is the set of skills and patterns of work adapted to immediate experience, instantaneous feedback loops, and deterministic outcomes. Executive mode is the set of skills and patterns of work adapted to operating with abstractions, delayed feedback loops, and mostly probabilistic outcomes.

If you’re a craftsman learning to operate in executive mode, everything you study will fall into one of three broad categories– the mundane, the mystical, and the meta. This is important because unlike craftsman mode, you study executive mode breadth first. For example, you learn to program by picking a programming language, then going deep until you master it. But you don’t get good at executive mode by learning everything there is to know about Steve Jobs.

So you have to approach studying executive mode breadth first. And to do that, you need to understand what the categories are.

First there is the mundane. The mundane is conventional operations techniques used in modern organizations– roughly everything you can outsource to a competent, well-incentivized professional CEO and expect good outcomes. Some of these techniques, like productivity and management, are applicable to executive mode in general. Others are domain specific. Here is an example from Jason Lemkin’s SaaStr:

You have to train your team in SaaS. How to go upmarket. How to drive up NPS. How to increase pricing. How to sell more of a solution. How to compete. How to get more referrals, and increase revenue retention.

The mundane doesn’t mean unimportant. A lot of people screw this up– it’s very, very important. It’s the nuts and bolts of the operation. If you want to work in executive mode, you have to get good at least at some of these things. To do that you study this: 1 2 3 4.

Then there is the mystical. The mystical goes something like this. Yes, the mundane is very important, and yes, conventional wisdom will get you far. But nobody seriously believes that tips on structuring your calendar, going upmarket or driving NPS will get you smart phones, or internet search, or personal computers. That kind of innovation is something else entirely. We know that given the right conditions entrepreneurs will create these products, but we only have a rough idea of what the conditions are. We don’t know yet how to reliably teach or replicate this.

So you read about exceptional projects and people, hoping to learn by osmosis. Reading the mystical is like reading history, or classics. You don’t read Symposium because the information is applicable to your life. You read it to learn to think like Socrates did. The mystical includes biographies, case studies, memos and even lore: 1 2 3 4. When people take the mystical too far, they cargo cult Mark Zuckerberg’s flip-flops, or Steve Jobs’s acid trips.

Finally there is the meta. The meta is an attempt to extract useful models from the mundane and the mystical to try and teach exceptionalism (or at least explain it post factum). Reading the meta won’t teach you how to hire a head of engineering or how open-source business models work; but it will give you tools at a higher level of abstraction to reason through these problems on your own. For example, how do you decide when to follow conventional wisdom and when to be contrarian? One possible answer to this question is to prefer contrarian questions. That’s one example of meta. Here are a few more: 1 2 3 4.

Most people, when they begin studying executive mode, gravitate toward one of the three categories to the exclusion of everything else. If you tend pragmatic you’ll likely restrict yourself to the mundane– what use is knowledge that doesn’t immediately help build products and close deals? If you tend intellectual you’ll likely restrict yourself to the meta– why read about a million details when you can learn the rules and figure out the details yourself from first principles?

That’s a mistake. Each category teaches you different kinds of things.

You can’t figure out the mundane from first principles any more than you can figure out how to be a chess grandmaster from learning the rules of chess. There is too much to learn, and not enough time. You need the knowledge others have painstakingly developed to go about your daily work. The mystical teaches you to think big, and gives you the courage to follow through on these thoughts. Humans are unusually good at learning by imitation; the mystical gives you the opportunity to imitate the best of humanity– people you may not have access to, or who may not even be alive. The meta enhances your ability to understand and to think independently. It gives you the ability to manipulate your idiosyncratic situation using higher order intellectual tools.

If we flatten the categories, there is productivity, management, domain-specific tactics, biographies, case studies, memos, lore, and meta reasoning. At the beginning– especially at the beginning– sample from them uniformly. Or at the very least, don’t neglect any one group.

Craftsman, executive— a tale of two modes

2019-11-25T00:00:00+00:00

Imagine being a shoemaker in a small village at the turn of the 20th century. You would know a great deal about running a business– how to procure materials, use tools, build relationships, price products, negotiate, hire, fire, learn your customers’s tastes and make products they want to buy. But you’re not the usual shoemaker. You feel the zeal of ambition and decide to expand. What will you discover when you start selling shoes outside the small radius of your village?

First, you will find that much of what you think about is now outside of your immediate experience. Before, when you procured materials for your workshop, made the shoes and sold them to your neighbors, you were physically present at each step of every transaction. You could touch the leather, measure your neighbor’s feet, examine the finished shoe, and shake hands at the end of the exchange. Everything about your process was immediate and tactile.

But as you start selling shoes in multiple villages, you will need to transition to working with abstractions. You won’t be able to meet all your customers or be present at every store. If you are really successful, eventually you may not know all your employees by name. You will have to start thinking in terms of market segments, demographics, store locations, cash flow and income statements. You will no longer be able to run large parts of your business through direct experience– you will have to think in abstract concepts and fly by instruments.

Second, your decisions will no longer be subject to immediate feedback. Before expanding, your could learn from your actions immediately (or at least very quickly). When you made a shoe you found out in a few days whether it sold. When you hired an assistant, you learned in a few weeks if they are a good employee.

When you make a shoe in your larger operation, it will take weeks before you find out if it sells. If you increase the size of your staff, it will be months before everyone is hired and trained. It may be even longer before you learn whether hiring more employees improved your organization. At almost every step the feedback loop is now much longer– and so it takes much longer to ascertain the outcome of your decisions. And it takes proportionally longer to learn and adapt.

Third, many of your outcomes will now be probabilistic rather than deterministic. For example, before the expansion the color brown either sold or it did not. But in the expanded world you are subject to the whims of forces you cannot predict or control. Maybe a new competitor flooded the market with cheap brown shoes; maybe brown is out of fashion this season in big cities where consumers are much more trend-conscious; maybe people just got sick of brown.

When you sold shoes in a small village your decisions either worked or they did not. In the expanded world of selling to a bigger, more complex market, a decision that worked yesterday may be a disaster today, and may work again tomorrow. You can no longer rely on determinism– you have to adapt to the new fact that many of the outcomes are now unpredictable, and that the way to model them is through random variables rather than first order logic.

Craftsman mode is the set of skills and patterns of work adapted to immediate experience, instantaneous feedback loops, and deterministic outcomes. Executive mode is the set of skills and patterns of work adapted to operating with abstractions, delayed feedback loops, and mostly probabilistic outcomes.

Many people are good at operating in craftsman mode as it’s been the default mode for most of human history. Fewer are good at operating in executive mode. It’s difficult to learn by osmosis as the density of good role models is still low, and we haven’t yet developed a systematic way of teaching it. And of course very few are good at what’s likely the most prized ability of all– switching between the two modes at will.

Nobody knows anything

2019-11-21T00:00:00+00:00

The following is an excerpt from William Goldman’s Adventures in the Screen Trade. Goldman was as successful a screenwriter as anybody. So this excerpt is worth paying close attention to.

Not one person in the entire motion picture field knows for a certainty what’s going to work. Every time out it’s a guess— and, if you’re lucky, an educated one.

They don’t know when the movie is finished: B. J. Thomas’s people, after the first sneak of Butch, were upset about their client’s getting involved with the song “Raindrops Keep Fallin’ on My Head.” One of them was heard to say, more than once, “B. J. really hurt himself with this one.” The initial preview of Star! was such a success that Richard Zanuck cancelled any further previews and sent a wire to his father, Darryl, that said, “We’re home. Better than Sound of Music.”

The Sound of Music was then the most popular movie in history, and Star! went on to become the Edsel of 20th Century-Fox: No matter how they readvertised it or changed the logo or the title, no one came. And Richard Zanuck has as keen a mind about commercial films as anyone.

They don’t know when the movie is starting to shoot either. David Brown, Zanuck’s partner, has said, “We didn’t know whether Jaws would work, but we didn’t have any doubts about The Island. It had to be a smash. Everything worked. The screenplay worked. Every actor we sent it to said yes. I didn’t know until a few days after we opened and I was in a bookstore and I ran into Lew Wasserman and said ‘How’re we doing?’ and he said, ‘David, they don’t want to see the picture.’”

They don’t want to see the picture– maybe the most chilling phrase in the industry.

Now, if the best people around don’t know at sneaks, and they don’t know during shooting, you better believe that executives don’t know when they’re trying to give a thumbs-up or down; they’re trying to predict public taste three years ahead and it’s just not possible.

Obviously, I’m asking you to take my word on this and there’s no reason really that you should, because pictures such as Raiders of the Lost Ark probably come to mind. Which, I grant, was an unusual film.

Raiders is the number-four film in history as this is being written. I don’t remember any movie that had such power going in. It was more or less the brainchild of George Lucas and was directed by Steven Spielberg, the two unquestioned wunderkinder of show business (Star Wars, Jaws, etc.). Probably you all knew that. But did you know that Raiders of the Lost Ark was offered to every single studio in town— and they all turned it down?

All except Paramount.

Why did Paramount say yes? Because nobody knows anything. And why did all the other studios say no? Because nobody knows anything. And why did Universal, the mightiest studio of all, pass on Star Wars, a decision that just may cost them, when all the sequels and spinoffs and toy money and book money and video-game money are totaled, over a billion dollars? Because nobody, nobody-not now, not ever—knows the least goddam thing about what is or isn’t going to work at the box office.

One additional anguish executives must cope with is that hot streaks don’t last. A recent newspaper article mentioned how the other studios were gloating over what was happening at Columbia.

Columbia had been sizzling, but then Annie went wildly over budget. And an expensive action film wouldn’t cut together coherently. And everybody knew that the set of Tootsie was not where you wanted to spend your summer vacation. And they had passed on E.T.

Columbia had had it, developed it for a million dollars, took a survey, and discovered the audience for the movie would be too limited to make it profitable. So they let it go. (Universal picked it up and may make back the billion they didn’t earn by dropping Star Wars.)

David Picker, a fine studio executive for many years, once said something to this effect: “If I had said yes to all the projects I turned down, and no to all the ones I took, it would have worked out about the same.”

In any case, do not send to know why studio executives have insomnia. It goes with the territory….

Startup idea checklist

2019-03-26T00:00:00+00:00

I’ve been tinkering with different startup ideas and needed a good checklist to think through them. There are great templates for this already: The YC application, Amazon’s internal press release, and Sequoia’s Writing a Business Plan. I found myself mixing and tweaking these templates because they don’t exactly match my model of the world, so I wrote up my own list.

I use this list both to develop ideas and filter them. If you adopt it, be careful about using it as a filter. Remember that in the early stages, good ideas are very easy to kill.

Product

Who are the users?
<= 70 chars
Don Valentine: “Who cares?”
What is the essence of their dissatisfaction? If they read this answer, would they say “thanks, I wish I’d thought of putting it that way”?
<= 240 chars
@benedictevans: “The iTunes Store solved a user problem. So did the App Store. And so did Spotify and Apple Music, and indeed Apple News. But what user problem is solved by Apple’s commissioning TV shows?”
What are you building for them?
<= 70 chars
Peter Drucker: Is the product being designed for the customer, or at the customer?
Write a tweet from a hypothetical customer explaining the product and how it eliminates their dissatisfaction.
@BrianNorgard: “No one cares about your product. Who built it, its features, the origin story — it’s all superfluous. People only find value in what your product can do for them right now. Save people time. Save people money. Give people an escape. The selfish hand will always govern.”
Peter Drucker: Are you really doing the best you can to help the customer?
Write a blog post title for your product launch. Is it surprising? Is it new? Will your target customers want to click on it? Will they want to share the link? Will they still share it the next day?
<= 70 chars
Write the first paragraph of your product announcement blog post. Include the product name, an explanation of what the product is, the target market, the main benefit, and the call to action.
<= 240 chars
What “metrics of goodness” do your target customers care about? Does your product dominate every available alternative on these metrics? (i.e. what can you do that no one else can do?)
<= 240 chars
See also: The Rise of Worse is Better, Worse is worse, Jeff Bezos explains Amazon.com
Is your product as awesome as it could be? Probably not.
y/N
See also: Elon Musk Advises CEOs

Growth

Fill in the bottom-up market size equation: NUM_USERS * ACV = MARKET_SIZE. Are your numbers credible? Find a good reference class if you’re building something completely new.
See also: Shut up and multiply
Which subset of your target customers are so constrained by the status quo, they’ll welcome a buggy product?
<= 140 chars
List your first ten customers.
<= 240 chars
See also: Do Things that Don’t Scale
Which playbook will you use to get customers after the first ten?
<= 240 chars
See also: Five ways to build a $100 million business
What would need to be true in 18 months for you to get essentially unlimited cheap capital? How will you achieve that?
<= 240 chars
See also: Fast

Strategy

Why now? What’s true about the world that nobody else figured out yet?
<= 240 chars
What is the most ambitious achievable milestone for your company within a 25 year time horizon?
<= 70 chars
Is your product a credible advance toward this milestone?
Yes/no
What’s the next credible advance toward this milestone? The one after that? The one after that?
<= 240 chars
See also: Tesla master plan, iPhone runs OSX
How will you build a moat?
<= 240 chars
See also: How Competitive Forces Shape Strategy

Meaning

What would reaching your 25 year milestone mean for the world? Is this future really exciting? How many years of your life would you give up to teleport there? If you found yourself in this counterfactual world, would you want to go back?
<= 140 chars
See also: America Online commercial in 1995, Steve Jobs presents WebObjects at MSPDC 1996
If another company was working on this idea and not you, what would you think about it? Would you join them?
Yes/no
Imagine yourself standing in front of your team, investors, family, and friends. You’ve failed, and they’re waiting for you to speak. What will you say? Are you willing to work on this problem given that failure is the default?
<= 480 chars
See also: Your intervention won’t work

Bonus

What’s your company’s stock ticker symbol?
@sama: “it’s easy/fun to say every new startup you hear about is bad. you will usually be right. you will never be successful.”
Is it likely to be the most important company started this decade?
George Orwell: “Whoever is winning at the moment will always seem to be invincible.”
See also: YC top companies

Thanks to Darryl Ramm for feedback on this post.

RethinkDB: why we failed

2017-01-18T00:00:00+00:00

When we announced that RethinkDB is shutting down, I promised to write a post-mortem. I took some time to process the experience, and I can now write about it clearly.

In the HN discussion thread people proposed many reasons for why RethinkDB failed, from inexplicable perversity of human nature and clever machinations of MongoDB’s marketing people, to failure to build an experienced go-to-market team, to lack of numeric type support beyond 64-bit float. I aggregated the comments into a list of proposed failure reasons here.

Some of these reasons have a ring of truth to them, but they’re symptoms rather than causes. For example, saying that we failed to monetize is tautological. It doesn’t illuminate the reasons for why we failed.

In hindsight, two things went wrong – we picked a terrible market and optimized the product for the wrong metrics of goodness. Each mistake likely cut RethinkDB’s valuation by one to two orders of magnitude. So if we got either of these right, RethinkDB would have been the size of MongoDB, and if we got both of them right, we eventually could have been the size of Red Hat[1].

Terrible market

Our thinking went something like this. New companies aren’t getting built on top of Oracle, so there is a window of opportunity to build a new infrastructure company. The database market is huge. If we build a product that captures some of that market, we’ll end up building a very successful company.

Unfortunately you’re not in the market you think you’re in – you’re in the market your users think you’re in. And our users clearly thought of us as an open-source developer tools company, because that’s what we really were. Which turned out to be very unfortunate, because the open-source developer tools market is one of the worst markets one could possibly end up in. Thousands of people used RethinkDB, often in business contexts, but most were willing to pay less for the lifetime of usage than the price of a single Starbucks coffee (which is to say, they weren’t willing to pay anything at all).

This wasn’t because the product was so good people didn’t need to pay for support, or because developers don’t control budgets, or because of failure of capitalism. The answer is basic microeconomics. Developers love building developer tools, often for free. So while there is massive demand, the supply vastly outstrips it. This drives the number of alternatives up, and the prices down to zero.

To see how this plays out for other companies consider MongoDB (valued at roughly $1.6B with ~700 employees), and Docker (valued at roughly $1B with ~300 employees). Both companies completely dominate in their respective markets. Two very rough rules of thumb for private growth stage technology companies is that valuations are a 10x multiple of annual revenue, and that revenue per employee is around $200K/year. Which means that MongoDB’s annual revenue is around $140-$160M, and Docker’s annual revenue is around $60-$100M.

That looks pretty good, until you look at dominant B2B technology companies in markets that aren’t developer tools. Companies like SalesForce, or Palantir, or Box (which faces stiff competition). All of a sudden MongoDB and Docker start looking tiny.

And these are massive successes. If relatively established companies with existing partnerships, distribution infrastructure, and access to large accounts are having trouble growing, what does it mean for a startup in its germination stage?

For us, it meant an intractable customer acquisition funnel. If a startup in a fertile B2B market has to process a hundred leads to get to ten opportunities to get to a single sale, for a developer tools startup that number goes up 10x. You have access to plenty of high quality prospects – lots of people are downloading your product and engaging with you, but you have to burn through a ridiculous number of leads to converge to a single sale.

This has disastrous domino effects. It demoralizes the team, and makes it very challenging to attract investment and hire top talent. In turn, that constrains your resources so you can’t make sufficient investment in product and distribution. Startups live and die by momentum, and early distribution challenges almost always doom you to eventual death.

Wrong metrics of goodness

Ok, so the market is bad, but other developer tools companies are still selling a lot of product. Why not RethinkDB?

While we couldn’t do anything about the dynamics of the market (other than building something else), the product decisions were entirely within our control. We wanted to build an elegant, robust, and beautiful product, so we optimized for the following metrics:

Correctness. We made very strict guarantees, and fulfilled them religiously.
Simplicity of the interface. We took on most of the complexity in the implementation, so application developers wouldn’t have to.
Consistency. We made everything from the query language, to the client drivers, to cluster configuration, to documentation, to the marketing copy on the front page as consistent as possible.

If these trade-offs seem familiar, they’re straight from the worse is better essay. It turned out that correctness, simplicity of the interface, and consistency are the wrong metrics of goodness for most users. The majority of users wanted these three trade-offs instead:

Timely arrival. They wanted the product to actually exist when they needed it, not three years later.
Palpable speed. People wanted RethinkDB to be fast on workloads they actually tried, rather than “real world” workloads we suggested. For example, they’d write quick scripts to measure how long it takes to insert ten thousand documents without ever reading them back. MongoDB mastered these workloads brilliantly, while we fought the losing battle of educating the market.
A use case. We set out to build a good database system, but users wanted a good way to do X (e.g. a good way to store JSON documents from hapi, a good way to store and analyze logs, a good way to create reports, etc.)

It’s not that we didn’t try to ship quickly, make RethinkDB fast, and build the ecosystem around it to make doing useful work easy. We did. But correct, simple, and consistent software takes a very long time to build. That put us three years behind the market.

By the time we felt RethinkDB satisfied our design goals and we were confident enough to recommend it to be used in production, almost everyone was asking “how is RethinkDB different from MongoDB?” We worked hard to explain why correctness, simplicity, and consistency are important, but ultimately these weren’t the metrics of goodness that mattered to most users.

To be honest, it hurt. It hurt a lot. It was unfathomable to us why people would choose a system that barely does the thing it’s supposed to do (store data), has a big kernel lock, throws away errors at random, implements single node features that stop working when you shard, has a barely working sharding system despite it being one of the core features of the product, provides essentially no correctness guarantees, and exposes a hodge-podge of interfaces that have no discernible consistency or unity of vision.

Every time MongoDB shipped a new release and people congratulated them on making improvements, I felt pangs of resentment. They’d announce they fixed the BKL, but really they’d get the granularity level down from a database to a collection. They’d add more operations, but instead of a composable interface that fits with the rest of the system, they’d simply bolt on one-off commands. They’d make sharding improvements, but it was obvious they were unwilling or unable to make even rudimentary data consistency guarantees.

But over time I learned to appreciate the wisdom of the crowds. MongoDB turned regular developers into heroes when people needed it, not years after the fact. It made data storage fast, and let people ship products quickly. And over time, MongoDB grew up. One by one, they fixed the issues with the architecture, and now it is an excellent product. It may not be as beautiful as we would have wanted, but it does the job, and it does it well.

When it became clear in mid-2014 that we couldn’t compete, we worked hard to differentiate from MongoDB. We found a very elegant way to add realtime push, hoping to enable developers to build a generation of apps they couldn’t build before. But that wasn’t enough. Suddenly we found ourselves competing with Meteor and Firebase, companies that were dedicated to solving the realtime problem for years before we even thought of it. Again we were three years behind the market, and again we found ourselves unable to compete.

What about the cloud?

A few people suggested that we should have built a cloud offering. We actually did have one in the works, so it’s an interesting topic I’d like to cover.

The obvious problem with a small database company building a cloud service is that it pattern matches to a common startup failure mode – splitting focus. Building, shipping, and operating reliable multi-tenant cloud services is hard. It requires non-trivial expertise and resources, so if you go down that path you find yourself running two startups at once. But we were facing an existential threat and were rapidly running out of options, so we gave it a shot anyway. Let’s suppose for the moment we could have pulled it off.

Our reasoning went like this. A database cloud offering could mean one of three things: managed hosting, database as a service (DBaaS), or value-added platform as a service (PaaS). Let’s do a quick back of the napkin market analysis using a $200K/employee in annual revenue rule of thumb we used above:

	Managed Hosting	DBaaS	PaaS
Company	Compose.io, mLab	FaunaDB	Parse, Firebase, Meteor
Employees	~30	~30	~30
Revenue	< $10M	< $10M	< $10M

So these markets are small, even smaller than the database market itself. But could one of them be a better bet than others?

Managed hosting is essentially running the database for people on AWS so they don’t have to. The alternative to using these services is setting up the database on AWS yourself. That’s a pain, but it isn’t actually that hard. So there is a very hard cap on how much managed database hosting services can charge. Considering that Compose.io and mLab are offering MongoDB which has one to two orders of magnitude more users than RethinkDB, we reasoned that offering managed hosting wouldn’t make a dent.

Database as a service is a more complex version of managed hosting – DBaaS offerings abstract node management from the user entirely. You simply run your queries and the system handles them. You don’t know anything about how many nodes are run under the hood. This business is very challenging – partly because DBaaS companies have to compete with the giants (e.g. DynamoDB and DocumentDB), and partly because customers are very resistant to completely hand off data management to a startup when there are so many other substitutes and alternatives (do you know anyone who uses a DBaaS offering from a startup?) So a DBaaS offering was out.

The last option was to build a value-added platform as a service. We thought this was a promising direction because here we had a massive technical advantage. Firebase and Meteor had to build application-level realtime logic on top of MongoDB, which fundamentally limits the realtime querying capabilities and performance at scale. On the other hand, we controlled the stack all the way down, so we could offer significant advantages Firebase and Meteor couldn’t build.

So we built Horizon and started working on Horizon Cloud – a way for users to deploy and scale RethinkDB/Horizon apps. The challenges of building three large projects (RethinkDB, Horizon, and Horizon Cloud) with a very small team eventually caught up with us, and we never managed to ship the cloud offering before we ran out of money. Kudos to the engineering team, though. They came very, very close.

Meta questions

There is one more level of root cause analysis that we can do. Why did we pick a bad market and optimize the product for the wrong metrics?

When I was a little kid I wanted to build my own radio. I made a box out of plywood, threw some metal junk inside, and connected the box to a power cord. I had books on electronics at home, but didn’t think I needed them – I had unwavering faith that I could do it on my own. Eventually I did build a working receiver, but it took me years to finally realize I needed to learn basic electronics.

Early RethinkDB was quite a bit like that. We had no intuition for products or markets, so we’d go through the motions of building a company without actually understanding what we were doing. What’s more, we had enormous optimism bias. Just like physicians know that gifts from pharmaceutical companies have biasing effects for other physicians but believe they are immune from the effect, we believed we were immune from the laws of economics and the math of running a business. The math, of course, eventually caught up with us.

Could we have done anything to avoid these mistakes? Not any more than I could have built a working radio as a little kid. We were unconsciously incompetent, and it took years for that incompetence to become conscious.

A few people pointed out that we would have done better if we had built an experienced go-to-market team. That’s 100% true, but the timing of our personal development didn’t line up with the needs of the company. Initially we didn’t know we needed go-to-market expertise, so we didn’t seek to include it on the founding team[2]. By the time we built up a mental model that maps well to reality, we found ourselves short on cash, in a difficult market filled with capable competitors, and a product that’s three years behind. By then, the best go-to-market team in the world couldn’t have saved us.

Parting thoughts

Many people have very strong feelings about the developer tools market. Engineers love building developer tools, so they badly want developer tools companies to thrive.

I am hesitant to dismiss the market entirely – partly because I don’t want to generalize from a single experience, partly because I don’t like saying “it cannot be done”, and partly because there are quite a few exceptions. GitHub, MongoDB, and Docker have built formidable companies. GitLab and Unity seem to be doing well.

If you do set out to build a developer tools company, tread carefully. The market is filled with good alternatives. User expectations are high and prices are low. Think deeply about the value you’re offering to the customer. Remember – wanting the world to be a certain way doesn’t make it so.

In 2009, we were pitching the early idea for RethinkDB (we had no software yet) to an audience of investors at the YCombinator demo day. We ended the pitch with a slide of three key points to remember. “If you only remember three things about RethinkDB,” we said, “remember these.” It worked. People didn’t remember anything else about the pitch, but they did remember the three points at the end.

I’ll now leave you with three key points to remember. If you remember anything about this post, remember these:

Pick a large market but build for specific users.
Learn to recognize the talents you’re missing, then work like hell to get them on your team.
Read The Economist religiously. It will make you better faster.

[1] Don’t read into these numbers too closely. I’m ball-parking it, but it should give you a general idea of the cost of these mistakes.

[2] Incidentally, recognizing good business people without having strong business intuition is about as hard as recognizing good engineers without having a strong intuition for engineering.

Mental models

2016-12-22T00:00:00+00:00

These are some mental models I find useful. They’re rooted in decades of experience of thousands of experts – a modern equivalent of folk wisdom. Mental models are useful to quickly and correctly reason about seemingly intractable problems. They require quite a bit of intuition to properly internalize, but once you’ve internalized them they’re relatively easy to apply. They’re also easy to forget in the moment – use this post as a checklist when thinking about complex problems.

This is a living document. Instead of creating an exhaustive list on day one, I will add models as they arise (and as I discover new ones).

Productivity

The small-improvements method – the observation that psychologically frequently making small incremental improvements is a better approach than attempting to fix big looming problems once.
The just-get-started method – Joel Spolsky’s observation that just starting to work on a small, concrete, finishable problem puts your consciousness in a productive state.
Corollary: Just do something concrete. Anything. Do your laundry, or dust the counters, or add a single unit test. Just do something.
The top-five-problems method – Richard Hamming’s algorithm for doing important work. Periodically ask yourself: “what are the top five most important problems in my field (and life), and why am I not working on them?”
Corollary: What are the top five most important problems in your field (and life), and why aren’t you working on them?
The LRU prioritization method – since you can only work on one problem at a time, it’s usually sufficient to pick the most important problem, work on that, and ignore everything else. This method also works with organizing most things (from email to physical possessions).
The teaching method – Richard Feynman’s observation that teaching the basics is an excellent method for generating profound new ideas, and for putting consciousness in a productive state.
Corollary: If you’re stuck, put yourself in a position where you have to teach someone the basics.
Planning fallacy – the observation that humans are overly optimistic when predicting success of their undertakings. Empirically, the average case turns out to be worse than the worst case human estimate.
Corollary: Be really pessimistic when estimating. Assume the average case will be slightly worse than the hypothetical worst case.
Corollary: When estimating time, upgrade the units and double the estimate (e.g. convert “one week” to “two months”).
Forcing function – an external, usually social, constraint that increases the probability of accomplishing a set of tasks.
Example: Pair programming.

Hypothesis evaluation

Efficient market hypothesis – the state of any given issue in the world is roughly as close to optimal as is currently possible.
Corollary: It’s unlikely that the status quo can be easily improved without significant resources.
Example: Cucumber juice probably doesn’t cure cancer.
Example: The iPhone app you wrote in a weekend probably doesn’t double the phone’s battery life.
Statistical mechanics – probabalistic systems that follow certain laws in the long run can have perturbations that diverge from these laws in the short run.
Corollary: Occasionally the status quo can be easily improved without significant resources (but it is unlikely that you found such an occassion).
Idiom: In the short run the market is a voting machine, but in the long run it is a weighing machine.
Idiom: If an economist saw a $100 bill on a sidewalk they wouldn’t pick it up (because if it were real, it would have been picked up already).
Base rates – you can approximate the likelihood of a specific event occurring by examining the wider probability distribution of similar events.
Example: You’re evaluating the probability of success of a given startup. Ask yourself, if you saw ten similar startups a year, how many of them are likely to succeed?
Example: You caught an employee stealing, but they claim they need money to buy medication and it’s the first time they’ve ever stolen anything. Ask yourself, if you saw ten employee thefts a year, how many of them are likely to be first offences?
Note: This method is especially useful to combat optimism and overconfidence biases, or when evaluating outcomes of events you’re emotionally close to.
See also: Techniques for probability estimates, reference class forecasting, prior probability.
Emic vs etic (aka inside vs outside view) – two perspectives you can choose when evaluating persuasive arguments. The inside view is time consuming and requires you to engage with the arguments on their merits. The outside view only requires you ask “what kind of person does sincerely believing this stuff turn you into?”
Corollary: You can usually predict correctness of arguments by evaluating superficial attributes of the people making them.
Example: If someone is wearing funny clothes, purports to know the one true way, and keeps talking about the glorious leader, you can usually dismiss their arguments without deeper examination.
Warning: This method usually works because most kooky people aren’t innovators, but will misfire in important situations because many innovators initially seem kooky.

Decision making

Inversion – the observation that many hard problems are best solved when they’re addressed backward. In other words figure out what you don’t want, avoid it, and you’ll get what you do want.
Corollary: Find out how people commonly fail doing what you do, and avoid failing like them.
Example: If you want to help India, ask “what is doing the worst damage in India and how can we avoid it?”
See also: Failure mode.
Bias for action – in daily life many important decisions are easily reversible. It’s not enough to have information – it’s crucial to move quickly and recover if you were wrong, than to deliberate indefinitely.
Idiom: One test is worth a thousand expert opinions.
Idiom: The best thing you can do is the right thing, the next best thing is the wrong thing, and the worst thing you can do is nothing.
Note: The best people do this naturally, without brooding, and with a light touch.
Expected value – a simple model for evaluating uncertain events (multiply the probability of the event by its value).
Corollary: Sometimes you’ll have to estimate probabilities when it feels really hard to do.
Example: Chance of winning NY lotto is 1 in 292,201,338 per game. Let’s say the grand prize is $150M and ticket price is $1. Then the expected value is roughly $0.5. Since $0.5 < $1, the model tells us the game isn’t worth playing.
Warning: Looking at expected value often isn’t enough. You need to consider utility to make good decisions.
See also: Techniques for probability estimates, shut up and multiply, scope insensitivity.
Marginal utility – the change in utility from the change in consumption of a good. Marginal utility usually diminishes with increase in consumption.
Example: The first car in your garage improves your life significantly more than the second one.
Example: Because utility loss from losing a dollar is negligible relative to utility gain from winning NY Lotto at ridiculously low odds, it might be worth buying a ticket even at negative expected value (but seriously, don’t).
Corollary: Think through your utility function carefully.
Strategy and tactics – empirically decisions tend to fall into one of two categories. Strategic decisions have long-term, gradual, and subtle effects (they’re a gift that keeps on giving). Tactical decisions are encapsulated into outcomes that have relatively quick binary resolutions (success or failure).
Example: Picking a programming language is a strategic decision.
Example: Picking a line of reasoning when trying to close a sale is a tactical decision.
Corollary: Most people misuse these terms (e.g. “we need a strategy for this meeting”).

People

IQ, RQ, and EQ – respectively, intelligence quotient (assessment of the mind’s raw horse power), rationality quotient (assessment of how well the mind’s models map to the real world; a measure of efficiency of the IQ’s application to real problems), and emotional quotient (ability to recognize and label emotions).
Corollary: brilliant people can be jerks and kooks, empathic people can have wacky ideas about reality, and effective people can have average intelligence.
Structure and agency – the observation that human behavior derives from a balance of internalized cultural patterns and capacity to act independently. The interaction of these two properties influences and limits individual behavior.
Corollary: Pay attention to the need for structure and independence in each individual.
Corollary: Put a structure in front of even the most independent-minded people, and they’ll internalize it.
Corollary: People often behave the way they believe their role requires them to (as opposed to the actual requirements of the role).
Corollary: Pay attention to how people perceive their own roles, and break their expectations with caution.
Social status – the observation (particularly in improv) that social status is so important to humans, that modeling status alone results in extremely realistic performances.
Corollary: Pay attention to how people perceive their own status, and break their expectations with caution.
See also: Self-serving bias.
Controlled vulnerability: – the observation that humans are attracted to confidently expressed vulnerability in others but are scared to be vulnerable themselves.
Corollary: Humans feel strong attraction towards others who confidently display vulnerability.
Corollary: Humans feel a strong desire to reciprocate vulnerability. Vulnerability expression by others gives them a sense of safety to express themselves, followed by a feeling of relief and a strong bond with the counterpary.

Groups

Mere-exposure effect – an observation that humans tend to develop a preference for things, people, and processes merely because they are familiar with them. This effect is much stronger than it initially seems.
Corollary: Merely putting people in a room together repeatedly, giving them a shared direction, symbology, and competition will create a group with very strong bonds.
See also: In-group favoritism.

Communication

Story arc – human beings are wired to respond to storytelling. A story arc is a way to structure ideas to tap into this response, typically by describing a change in the world.
Example: Once upon a time there was ___. Every day, ___. One day ___. Because of that, ___. Because of that, ___. Until finally ___.
Writing well – use arresting imagery and tabulate your thoughts precisely. Never use a long word where a short one will do. If it’s possible to cut a word out, always cut it out. Don’t hedge – decide what you want to say and say it as vigorously as possible. Of all the places to go next, choose the most interesting.
Charitable interpretation – interpreting a speaker’s statements to be rational and, in the case of any argument, considering its best, strongest possible interpretation. Charitable interpretation makes conversations (and relationships) go better.
Nonviolent Communication (aka NVC) – a communication framework that allows expressing grievances and resolving conflicts in a non-confrontational way. Structuring difficult conversations as described in NVC makes the process dramatically less painful. NVC contains four components: (1) expressing facts, (2) expressing feelings, (3) expressing needs, and (4) making a request.
Example: You didn’t turn in the project yesterday. When that happened I felt betrayed. I need to be able to rely on you to have a productive relationship. In the future, could you notify me in advance if something like that happens?

Policy

Global utility maximization – our innate sense of fairness is often unsatisfiable, and attemping to satisfy it can occasionally cause much grief in exchange for little gain. It’s much better to optimize for the needs of the many, not for an idealistic notion of fairness.
Corollary: There are times when it makes sense to be unfair to the individual in the interest of the common good.
Example: It makes sense to fire an underperforming employee who has valid excuses for their poor performance.
Idiom: It is the greatest happiness of the greatest number that is the measure of right and wrong.
See also: Preference utilitarianism.
Tragedy of the commons – a set of circumstances where individuals acting independently in a reasonable manner behave contrary to the common good.
Example: Tourists taking small artifacts from popular attractions.
Corollary: Governance is necessary to preserve the common good.
Front page test – an ethical standard for behavior that evaluates each action through the lens of the media/outside world.
Example: What would happen if HN found out we’re mining our users’s IMs?
Warning: Incentivizes extreme risk aversion, often without appropriate consideration for potential gain.
Reasonable person principle – a rule of thumb for group communication originated in CMU. It holds that reasonable people strike a suitable balance between their own immediate desires and the good of the community at large.
Corollary: Fire people that are offensive or easily offended. (It usually turns out that people who possess one of these qualities, possess both.)
Note: unreasonable persons can be extremely valuable in greater society (e.g. journalists, comedians, whistleblowers, etc.), but usually not in small organizations.
Overton window – the range of ideas a particular group of people will accept. Ideas range in degree of acceptance from policy, to popular, sensible, acceptable, radical, and unthinkable.
Corollary: you need to be sensitive to the overton window when presenting the group with cultural changes.
Political capital – the trust and influence a leader wields with other people. Political capital increases when you make other people successful and decreases when you make unpopular decisions.
Corollary: Spend political capital carefully.

Product design

Target market – a predicate that partitions new leads into opportunities and distractions. A good target market function is terse, has a discoverable domain, and has a well defined probability of close in a specific time bound.
Example: Anyone who has a Cisco password has a 50% probability of close within 30 days.
Internal press release – you start developing a product by writing an internal press release first, explaining to target customers why the product is useful and how it blows away the competition. You then test it against potential users (it’s much easier to iterate on the press release than the product).
Corollary: If the press release is hard to write, then the product is probably going to suck.
Quantum of utility – a rule of thumb for launching the product. A product possesses a quantum of utility when there is at least some set of users who would be excited to hear about it, because they can now do something they couldn’t do before.
Note: “Launch” can be defined as a private beta, or even giving the product to a friend. The point is to get it into the hands of someone who’s not in the building as soon as possible.
Worse is better – a design philosophy which states that solving the customer’s problem and leaving unpolished rough edges empirically outperforms “beautiful” products.
Example: Lisp Machines vs C/Unix.
See also: Worse is worse.
Kano model – a model for categorizing possible features to optimize resource allocation. Essentially partitions the product into gamechangers, showstoppers, and distractions.
See also: How to build great products.

Business

Five forces – a model for analyzing the competitive intensity and therefore attractiveness of an industry. The five forces are: threat of new entrants, threat of substitutes, bargaining power of buyers, bargaining power of suppliers, and industry rivalry.
Note: this is essentially a base rate estimation model for companies in an industry.
Power of defaults – the observation that people favor the familiar over novel places, people, things, and processes.
Corollary: Overcoming the familiarity heuristic at scale requires enormous activation energy unavailable to startups.
Corollary: It is dramatically easier to capture mindshare before people’s minds are made up, than to change their mind later.
See also: Default effect, path of least resistance, brand equity.
Economies of scale – the advantages due to size or scale of operation, where cost per unit decreases with increasing scale.
See also: Network effects, brand equity, first mover advantage.
Price/performance curve – the observation that the price of important technology drops and performance improves over time.
Example: Moore’s Law.