Uncanny Valley

by Anna Wiener

  Out of self-protection, I stuck to the narrative that I was moving across the country just to try something new. I had never even lived outside of the tristate area. San Francisco had a great music scene, I said, unconvincingly, to anyone who would listen. It had medical marijuana. Working in analytics would be an experiment in separating my professional life from my personal interests. The startup gig was just a day job, I claimed, something to support me while I was otherwise creatively productive. Maybe I would start the short-story collection I had always wanted to write. Maybe I would take up pottery. I could finally learn bass.

  It was easier, in any case, to fabricate a romantic narrative than admit that I was ambitious—that I wanted my life to pick up momentum, go faster.

  When I arrived back in San Francisco, with a fresh haircut and two fraying duffel bags, I felt intrepid and pioneering. I did not know that thousands of people had already headed west for a crack at the new American dream, that they had been doing so for years. I was, by many standards, late.

  It was a moment of corporate obsequiousness to young men. Tech companies were importing freshly graduated computer science majors from all over the world, putting them up in furnished apartments, paying their cable and internet and cell phone bills, and offering hundred-thousand-dollar signing bonuses as tokens of thanks. The programmers arrived with a flood of nontechnical carpetbaggers: former Ph.D. students and middle-school teachers, public defenders and chamber music singers, financial analysts and assembly-line operators, me.

  I had booked another bedroom using the home-sharing platform, this time in the South of Market neighborhood, several blocks from the office. The room was on the garden level of a duplex, adjacent to a concrete patio and accessed through an alley, just past the recycling bins. It was decorated with the same lightweight, self-assembly furniture as my friends’ bedrooms back in Brooklyn. The woman who operated the rental was an entrepreneur in the renewable-energy space and described herself as never home.

  A few small boxes of my books, bedding, and clothing were already at the analytics startup, stacked in a supply closet. I had been self-conscious about spending down the relocation stipend, wanting to save money for the company. A part of me worried that if I spent too much, the offer would be rescinded. I didn’t want my new manager to think I was frivolous. Others had expensed new furniture, meals, weeks of rent, but I didn’t know that. I was still operating according to publishing austerity.

  The home-sharing platform offered an aspirational fantasy that I appreciated. Across the world, people were squeezing out the last of strangers’ toothpaste, picking up strangers’ soap in the shower, wiping their noses on strangers’ pillowcases. There I was as I had always been, only sleeping in a stranger’s bed, fumbling to replace a stranger’s spring-loaded toilet-paper holder, ordering sweaters off a stranger’s Wi-Fi network. I liked examining someone else’s product selections, judging their clutter. I wasn’t thinking about how the home-sharing platform might also be driving up rents, displacing residents, or undermining the very authenticity that it purported to sell. Mostly, the fact that it functioned, and nobody had murdered me, seemed like a miracle.

  I had given myself a few days to get adjusted before starting the job. In the mornings, I bought coffee at a laundromat, consulted a crowdsourced reviewing app to find something to eat, and returned to my bedroom to spend the rest of the day reading technical documentation for the analytics software and panicking. The documentation was indecipherable to me. I didn’t know what an API was, or how to use one. I didn’t know how I would possibly provide technical support to engineers—I couldn’t even fake it.

  The night before my first day of work, too unmoored and overwhelmed to sleep, I scrolled through previous guests’ reviews of my room and realized that the apartment was owned by one of the founders of the home-sharing platform. I looked up the founder’s name and read an interview in which he detailed how designers could follow in his footsteps and become entrepreneurs. He called them “designpreneurs.” I watched a video of him delivering the keynote at a tech conference, breathing excitedly into the mic. I learned that he and his two cofounders had raised over a hundred million dollars, and investors were desperate to give them more.

  I looked around me at the blank walls, the closet door tilted on its hinges, the bars on the window, eager to identify hints of his success. But the designpreneur hadn’t slept in the room for years. He had moved into a gleaming, art-filled warehouse conversion close to his office. He’d left nothing behind.

  * * *

  The analytics startup made a pickax-during-the-Gold-Rush product, the kind venture capitalists loved to get behind. History saw the Gold Rush as a cautionary tale, but in Silicon Valley, people used its metaphors proudly, provided they were on the right side of things. Pickaxes were usually business-to-business products. Infrastructure, not services. Just as startups in New York were eager to build off their city’s existing cultural legacy, by creating services for media and finance—or, more commonly, creating sleek interfaces to sell things that would require more time, money, energy, or taste to buy elsewhere—the same was true of the Bay Area, where software engineers sought to usurp older technology companies by building tools for other software engineers.

  It was the era of big data, complex data sets facilitated by exponentially faster computer processing power and stored, fashionably, in the cloud. Big data encompassed industries: science, medicine, farming, education, policing, surveillance. The right findings could be golden, inspiring new products or revealing user psychology, or engendering ingenious, hypertargeted advertising campaigns.

  Not everyone knew what they needed from big data, but everyone knew that they needed it. Just the prospect incited lust in product managers, advertising executives, and stock-market speculators. Data collection and retention were unregulated. Investors salivated over predictive analytics, the lucrative potential of steroidal pattern-matching, and the prospect of bringing machine-learning algorithms to the masses—or, at least, to Fortune 500 companies. Transparency for the masses wasn’t ideal: better that the masses not see what companies in the data space had on them.

  The analytics startup wasn’t disrupting anything so much as unseating big-data incumbents: slow-moving corporate behemoths whose products were technically unsophisticated and bore distinctly nineties user interfaces. The startup not only enabled other companies to collect customized data on their users’ behavior without having to write much code or pay for storage, but it also offered ways to analyze that data in colorful, dynamic dashboards. The cofounders had prioritized aesthetics and hired two graphic designers off the bat: men with signature hairstyles and large followings on a social network for people who referred to themselves as creatives and got excited about things like font sizing and hero images. In general, it was hard to say what, exactly, the designers did all day, but the dashboards were both friendly and elegant. The software looked especially pleasing, trustworthy, airtight. Good interface design was like magic, or religion: it cultivated the mass suspension of disbelief.

  I had no qualms about disrupting extant corporations in the big-data space, no inherited nostalgia or fondness for business. I liked the underdog. I liked the idea of working for two kids younger than I was, who had dropped out of college and were upending the script for success. It was thrilling, in that sense, to see a couple of twentysomethings go up against middle-aged leaders of industry. It looked like they could win.

  * * *

  I was employee number twenty, and the fourth woman. Prior to my arrival, the Solutions team—four men, including the manager—had handled customer tickets themselves, attacking the support queue in shifts at the end of the workday, relaying the responsibility to avoid consecutive office-bound midnights. This strategy was effective for a while, but the user base was ballooning. The men couldn’t sustain the practice; they had their own jobs to do. They rearranged their desktop belongings and cleared a space for me.

  The men on the S
olutions team weren’t like the men from the e-book startup. They were weirder, wilder, funnier, harder to keep up with. They wore Australian work boots and flannel and durable, recycled polyester athletic vests, drank energy shots in the late afternoon and popped vitamin D in the mornings to stay focused and alert. They chewed powdered Swedish tobacco, packing it juicily behind their gums. Deep house and EDM leaked from their oversized headphones. At team gatherings they drank whiskey, neat, and, the following mornings, were prone to pounding a viscous liquid jacked up with electrolytes—sold as a remedy for small children with diarrhea—to flush away their hangovers. They had gone to top-tier private colleges and were fluent in the jargon of media studies and literary theory. They reminded me of my friends who had left San Francisco, but more adaptable and opportunistic, happier.

  The solutions manager assigned me an onboarding buddy, Noah, a curly-haired twenty-six-year-old with a forearm tattoo in Sanskrit and a wardrobe of workman’s jackets and soft fleeces. Noah was warm and loquacious, animated, handsome. He struck me as the kind of person who would invite women over to get stoned and look at art books and listen to Brian Eno, and then actually spend the night doing that. I had gone to college with men like this: men who would comfortably sit on the floor with their backs against the bed, men who self-identified as feminists and would never make the first move. I could immediately picture him making seitan stir-fry, suggesting a hike in the rain. Showing up in an emergency and thinking he knew exactly what to do. Noah spoke in absolutes and in the language of psychoanalysis, offering definitive narratives for everyone, everything. I had the uneasy feeling that he could persuade me to do anything: bike across America; join a cult.

  Noah and I spent my first few weeks in various corners of the office, carting around an overflowing bowl of trail mix and a rolling whiteboard, on which he patiently diagrammed how cookie tracking worked, how data was sent server-side, how to send an HTTP request, how to prevent a race condition. He was patient and encouraging, and made direct eye contact as we pushed through problem sets of hypothetical customer questions, various scenarios in which the software—or, more realistically, the user—had a meltdown.

  The product was actually deeply technical, though the company talked up its usability. The amount of information I needed to absorb, to be even marginally helpful to our customers, was intimidating. The learning curve looked unconquerable. Noah gave me homework and pep talks. He told me not to worry. Our teammates handed me beers in the late afternoon, and were confident and reassuring that I, too, would eventually scale up. I trusted them entirely.

  I was happy; I was learning. For the first time in my professional life, I was not responsible for making anyone coffee. Instead, I was solving problems. My job involved surveying strangers’ codebases and telling them where they’d gone wrong in integrating our product with theirs, and how to fix it. The first time I looked at a block of code and understood what was happening, I felt like nothing less than a genius.

  * * *

  It did not take long for me to understand the fetish for big data. Data sets were mesmerizing: digital streams of human behavior, answers to questions I didn’t know I had. There was more every second. Our servers, and the company’s bank account, absorbed this unstoppable wave.

  Our bread and butter was engagement: actions that demonstrated the ways users were interacting with a product. This was a turn away from the long-running industry standard, which prioritized metrics like page views and time on site, metrics that the CEO called bullshit. Engagement, he said, was distinguished from the bullshit because it was actionable. Engagement generated a feedback loop between the user and the company. User behavior could dictate product managers’ decisions. These insights would be fed back into an app or website, to dictate or predict subsequent user behavior.

  The software was flexible, intended to function as easily for fitness trackers or payment processors as for photo-editing and ride-sharing apps. It could be integrated into online boutiques, digital megamalls, banks, social networks, streaming and gaming websites. It gathered data for platforms that enabled people to book flights or hotels or restaurant reservations or wedding venues; platforms for buying a house or finding a house cleaner, ordering takeout or arranging a date. Engineers and data scientists and product managers would inject snippets of our code into their own codebases, specify which behaviors they wanted to track, and begin collecting data immediately. Anything an app or website’s users did—tap a button, take a photograph, send a payment, swipe right, enter text—could be recorded in real time, stored, aggregated, and analyzed in those beautiful dashboards. Whenever I explained it to friends, I sounded like a podcast ad.

  Depending on the metadata, users’ actions could be scrutinized down to the bone, at the most granular level imaginable. Data could be segmented by anything an app collected—age, gender, political affiliation, hair color, dietary restrictions, body weight, income bracket, favorite movies, education, kinks, proclivities—plus some IP-based defaults, like country, city, cell phone carrier, device type, and a unique device identification code. If women in Boise were using an exercise app primarily between the hours of nine and eleven in the morning—only once a month, mostly on Sunday, and for an average of twenty-nine minutes—the software could know. If people on a dating website were messaging everyone within walking distance who practiced yoga, trimmed their pubic hair, and were usually monogamous but looking for a threesome during a stint in New Orleans, the software could know that, too. All customers had to do was run a report; all they had to do was ask.

  We also offered a secondary product, a people-analytics tool, for which some customers paid extra. The people-analytics tool stored individual profiles of users on those customers’ platforms. These contained streams of personalized, searchable activity, as well as any identifying metadata. The point of this tool was to facilitate outreach based on behavior, and incentivize re-engagement. An e-commerce store could search its own database to see which men, exactly, were filling online shopping carts with razor blades and beard oil but never checking out, and send those men an email, offering discounts or simply a passive-aggressive reminder that it might be time to shave. A food-delivery app, upon registering that a user had ordered a paleo TV tray six nights in a row, might trigger an in-app pop-up suggesting she try a carbohydrate. An exercise app could identify that a user had stopped a workout at the burpee section and automatically send a push notification asking him if he was still alive.

  The tool was free to use to a certain threshold, after which data was metered. If our customer companies acquired more users of their own, their volume of data would increase, and their monthly invoices would spike accordingly. This meant the tool was inherently lucrative, because every company wanted to grow. The underlying assumption was that if our customers were bringing on more users, they should also be bringing in more money; that revenue and usage were linked.

  This turned out to be generous. Many startups didn’t have a revenue model to begin with, optimizing instead for market penetration. In these cases, venture capital served as a placeholder for profit: companies acquired more users without bringing in more money, as if they were simply an intermediary between us and their investors’ bank accounts. Our payment structure was straightforward, simple, canny. It would have been logical, too, if logic—or basic economics—had any governance over the venture-backed ecosystem.

  * * *

  To do my job effectively, I had to be able to see customers’ code, as well as their dashboards. This was true of anyone in a customer-facing role; it was almost impossible to solve users’ problems if the problems weren’t in front of us. The simplest way for the analytics startup to achieve this was by granting those of us on the Solutions team access to all of our customers’ data sets: to see the tool as if we were logged in to any given user’s account, to experience our product through their eyes.

  Some called this setting God Mode. It wasn’t our customers’ payment, contact, and organizati
onal information—though we could see that, too, if we needed to—but the actual data sets that they collected on their own users. This was a privileged vantage point from which to observe the tech industry, and we tried not to talk about it. “We’re not just selling jeans to miners,” Noah said. “We’re doing everyone’s laundry.”

  God Mode was a business education. Engagement metrics could tell the story of a startup’s entire life span. Startups rumored to be rocket ships sputtered to get off the ground. Gaming apps spiked and flamed out within weeks. The descent into obsolescence was almost always broken by cushions of venture capital, but we could see the direction things would go.

  We all knew that internal permissions, limiting what we could see of customer data sets, would come eventually. We also knew, at least for the time being, that it wasn’t a priority for our Engineering team. This level of employee access was normal for the industry—common for small, new startups whose engineers were overextended. Employees at ride-sharing startups, I’d heard, could search customers’ ride histories, tracking the travel patterns of celebrities and politicians. Even the social network everyone hated had its own version of God Mode: early employees had been granted access to users’ private activity and passwords. Permissioning was effectively a rite of passage. It was a concession to the demands of growth.