Chaos Monkeys: Obscene Fortune and Random Failure in Silicon Valley

by Antonio García Martínez

The great source of both the misery and disorders of human life, seems to arise from over-rating the difference between one permanent situation and another. Avarice over-rates the difference between poverty and riches: ambition, that between a private and a public station: vainglory, that between obscurity and extensive reputation. The person under the influence of any of those extravagant passions, is not only miserable in his actual situation, but is often disposed to disturb the peace of society, in order to arrive at that which he so foolishly admires. —Adam Smith, The Theory of Moral Sentiments

Commercial credit is the creation of modern times, and belongs, in its highest perfection, only to the most enlightened and best governed nations. It has raised armies, equipped navies, and, triumphing over the gross power of mere numbers, it has established national superiority on the foundation of intelligence, wealth, and well-directed industry. —Daniel Webster, US Senate speech, March 18, 1834*

“Credit” is the third-person singular conjugation of the present tense of the Latin verb credere, “to believe.” It’s the most exceptional and interesting thing in the financial world. Similar leaps of belief underlie every human transaction in life: Your wife might cheat on you, but you hope otherwise. The online store you paid may not ship you your goods, but you trust otherwise. Credit derivatives are just the explicit encapsulations of such beliefs, in financial and contractual form, for corporate entities. Unlike other financial securities, such as shares of IBM stock or oil futures, a credit derivative is not even some theoretical value of a tangible good. It’s the perceived value of a complete intangible, the perception of the probability of meeting some future obligation. People often asked me in the early days of my tech career how I had gone from Wall Street to ads technology. Such a person almost certainly knew nothing about either industry, or the answer would have been obvious. I did the same thing the whole time: putting a price on a human’s perception, be it of a General Motors bond or a pair of shoes coveted on Zappos. It’s the same difference either way; only the scale of the money pile changes.

To paraphrase the very quotable Silicon Valley venture capitalist Marc Andreessen, in the future there will be two types of jobs: people who tell computers what to do, and people who are told by computers what to do. Wall Street was merely the first inkling. The next place where this shift would be seen at whopping scale in terms of both money and technology (though I didn’t realize at the time) was in Internet advertising. And after that, it would hit transportation (Uber), hostelry (Airbnb), food delivery (Instacart), and so on. To take the theory further, computation would no longer fill some hard gap in a human workflow process, such as the calculators used by accountants. Humans would fill the hard gaps in a purely computer workflow process, like Uber’s drivers. But we’re getting ahead of ourselves. There’s an additional lesson here. This shift from humans to computers took place predominantly on the equity side of things. The debt side of the financial world, for various reasons, still traded in what amounted to open-outcry markets with humans talking to one another, whether through phones or instant messaging systems. It was capitalism at the speed a tongue can wag or hands can type. This was mostly because a company’s debt is complex and multifarious, and entities like General Motors have hundreds if not thousands of different types of debt floating around the world’s trading floors. Briefly, they are not what economists call “fungible,” meaning interchangeable the...

One week into my new Silicon Valley life, and the lesson was this: if you want to be a startup entrepreneur, get used to negotiating from positions of weakness. I’d soon have trickier situations to negotiate than convincing a cop to let me take a cab. And so will you if you play the startup game. The next morning, I wasn’t merely hungover, but was in fact still mildly drunk. The company all-hands meeting, wherein the entire company gathered to hear about new deals and employees, and generally to get pep-rallied by Murthy Nukala, the CEO, was scheduled for noon that day. I had to be there or risk having my coworkers file a missing persons report, as well as look like a pussy. My frazzled brain was slow to realize my car was still somewhere in San Mateo. One hundred and thirty dollars and too much sunlight later, I was standing beside my four-wheeled Bavarian steed at the scene of last night’s triumph over the rule of law, and fifteen minutes later I was an acceptable five minutes late for the all-hands. As I walked into the company-wide meeting, a murmur was heard from a corner of the assembled crowd, expressing either surprise or amusement at my being both alive and unincarcerated. The company rumor mill had been busy that morning. I probably looked as pickled and embalmed as I felt. Murthy launched into his weekly harangue. The wheels of capitalism ground ever on.

Whether it be brand marketers trumpeting the new BMW X5, game developers getting players to spend real money on virtual goods, or someone selling an online nursing degree, the only difference is the time frame in which those different goals occur—in other words, the time between attention and action. If the time frame is very short, like browsing for and buying a shirt at nordstroms.com, it’s called “direct response,” or “DR” advertising. If the time frame is very long, such as making you believe life is unlivable outside the pricey mantle of a Burberry coat, it’s called “brand advertising.” Note that the goal is the same in both: to make you buy shit you likely don’t need with money you likely don’t have. In the former case, the trail is easily trackable, as the “conversion” usually happens online, usually after clicking on the very ad you were served.* In the latter, the media employed is a multipronged strategy of Super Bowl ads, Internet advertising, postal mail, free keychains, and God knows what else. Also, the conversion happens way after the initial exposure to the media, and often offline and in a physical space, like at a car dealership. The tracking and attribution are much harder, due to both the manifold media used and the months or years gone by between the exposure and the sale. As such, brand advertising budgets, which are far larger than direct-response ones, are spent in embarrassingly large broadsides, barely targeted or tracked in any way. Now you know ...

Here’s something you may not know: every time you go to Facebook or ESPN.com or wherever, you’re unleashing a mad scramble of money, data, and pixels that involves undersea fiber-optic cables, the world’s best database technologies, and everything that is known about you by greedy strangers. Every. Single. Time. The magic of how this happens is called “real-time bidding” (RTB) exchanges, and we’ll get into the technical details before long. For now, imagine that every time you go to CNN.com, it’s as though a new sell order for one share in your brain is transmitted to a stock exchange. Picture it: individual quanta of human attention sold, bit by bit, like so many million shares of General Motors stock, billions of times a day. Remember Spear, Leeds & Kellogg, Goldman Sachs’s old-school brokerage acquisition, and its disappearing (or disappeared) traders? The company went from hundreds of traders and two programmers to twenty programmers and two traders in a few years. That same process was just starting in the media world circa 2009, and is right now, in 2016, kicking into high gear. As part of that shift, one of the final paroxysms of wasted effort at Adchemy was taking place precisely in the RTB space. An engineer named Matthew McEachen, one of Adchemy’s best, and I built an RTB bidding engine that talked to Google’s huge ad exchange, the figurative New York Stock Exchange of media, and submitted bids and ads at speeds of upwards of one hundred thousand requests per sec...

So what idea were we pitching exactly? The short answer: it didn’t matter. Like the canniest early-stage investors, YC cared a lot more about the team at this point than whatever improbable idea we might have. The latter was really there only to judge the former; it was valueless in itself. Don’t believe me? Think your idea is worth something? Go and try to sell it, and see what sort of price you’ll get for it. Ideas without implementation, or without an exceptional team to implement them, are like assholes and opinions: everyone’s got one. Incidentally, the fastest way you can indicate your level of startup naïveté to a VC (or to anybody in tech), is either by claiming you’re in “stealth”—that is, with an idea so secretly valuable you can’t disclose it—or by forcing someone to sign a nondisclosure agreement before you even discuss it. You may as well tattoo LOSER on your forehead instead, to save everyone the trouble. To quote one Valley sage, if your idea is any good, it won’t get stolen, you’ll have to jam it down people’s throats instead. For the record, though, here’s what we came up with: One of the great challenges in marketing is the divide between the online and the offline worlds. What you buy via the Internet and your digital persona is completely separate from what you buy in person, and the two data streams rarely cross. Imagine if an advertiser knew you bought SUV-sized packages of diapers at your local Walmart; how would it target different Internet ads at y...

Here’s some startup pedagogy for you: When confronted with any startup idea, ask yourself one simple question: How many miracles have to happen for this to succeed? If the answer is zero, you’re not looking at a startup, you’re just dealing with a regular business like a laundry or a trucking business. All you need is capital and minimal execution, and assuming a two-way market, you’ll make some profit. To be a startup, miracles need to happen. But a precise number of miracles. Most successful startups depend on one miracle only. For Airbnb, it was getting people to let strangers into their spare bedrooms and weekend cottages. This was a user-behavior miracle. For Google, it was creating an exponentially better search service than anything that had existed to date. This was a technical miracle. For Uber or Instacart, it was getting people to book and pay for real-world services via websites or phones. This was a consumer-workflow miracle. For Slack, it was getting people to work like they formerly chatted with their girlfriends. This is a business-workflow miracle. For the makers of most consumer apps (e.g., Instagram), the miracle was quite simple: getting users to use your app, and then to realize the financial value of your particular twist on a human brain interacting with keyboard or touchscreen. That was Facebook’s miracle, getting every college student in America to use its platform during its early years. While there was much technical know-how required in scaling ...

comme il faut.

In reality, though, it usually worked like this: A female candidate who will buzzkill your weekly happy hour? “Cultural fit.” A soft-spoken Indian or Chinese engineer, quietly competent but incapable of the hard-charging egotism that Americans almost universally wear like they do blue jeans? “Cultural fit.” Self-taught kid from some crappy college you’ve never heard of, without that glib sheen of effortless superiority you get out of Harvard or Stanford? “Cultural fit.” And so it goes. Shaffer’s machine-gun questioning and imperiousness had rattled me. I suspected that I had failed to pass his bar, and I needed to clear my head. The day had been nothing but a series of interrogations inside small, gray, rotten-smelling rooms. The Guantánamo vibe was fatiguing. Despite the NSA-level security on checking in and the way we were handed off like booby-trapped hot potatoes that no one could drop, nobody appeared for the next interview. Wining and dining evidently not in the offing, I wandered off and tried to find something to eat.

At that point in time, Gokul Rajaram was a legendary éminence grise in the ad-tech world. The so-called godfather of AdSense, Google’s secondary gold mine after AdWords, Gokul was a constant presence on the conference circuit, and an omnipresent adviser or investor in just about every advertising technology company worth talking about. He too had come to Facebook via a small acqui-hire, though really that had been just a career breather between his time at Google and his hiring at Facebook. University at the Indian Institute of Technology (IIT), followed by an American MBA, he was your standard-issue Indian techie, and probably that country’s most valuable export after steel and Tata Motors. “What’s the first thing you would change about Facebook Ads if we hired you?” There was about as much polish and prologue to Gokul as that of a North Korean diplomat. “I’d build a conversion-tracking system. It’s unbelievable you don’t have one yet.” A conversion-tracking system is software that tells you if an advertisement has worked in driving a conversion (or “sale” in marketing-speak), and lets you retweak your marketing campaigns based on performance. An ads system without conversion tracking is like a car without rearview mirrors; nay, it’s like a car without even rear or side windows. All you can see is forward, merrily driving along, not even understanding what’s behind you or what you just ran over. It’s a danger to yourself and others, and it was a sign of just how out-of-to...