Interviewed by Larry Au on May 3, 2024 for the Summer Newsletter

Benjamin Shestakofsky is Assistant Professor of Sociology at the University of Pennsylvania, where he is affiliated with AI at Wharton and the Center on Digital Culture and Society. He is the author of Behind the Startup: How Ventural Capital Shapes Work, Innovation, and Inequality (University of California Press, 2024).

Larry: Could you start by telling us a little bit about how you became interested in studying startups? And what ultimately brought you to the start-up AllDone and what was the process of gaining entry like?

Ben: I was living in New York City before I went to grad school, and a friend from college asked if I would work for his new startup. It turns out he had just convinced a wealthy angel investor to fund it. What almost immediately struck me was that my friend-turned-boss had no idea how to run a company. Why did a rich investor trust a 25-year-old with hundreds of thousands of dollars? My boss talked such a good game that he convinced a reporter to write a feature story about the company in the New York Times Magazine, which was published just as the website was launching. The website crashed immediately. It was a mess. So the gap between perception and reality, which is so important in the startup field, is something I observed firsthand as an employee.

As a grad student, I became interested in developing a sociological perspective on startups. I was fortunate to have a friend who brokered a meeting with one of AllDone’s co-founders. AllDone was a company that ran a digital platform connecting buyers and sellers of local services across hundreds of occupations, like plumbers, DJs, and house cleaners. I pitched him on working as an unpaid intern in exchange for research access. He ran it by the team and they agreed. At the time I was taking a participant-observation methods class and assumed that this project would culminate in my MA paper. But soon my supervisor at the startup asked if I would be willing to increase my workload, and my role turned into a part-time, paid position. Not long after that, the CEO asked me if I would consider a full-time position. I ended up withdrawing from grad school for a year so I could take on a role in middle management while continuing my research.

Larry: I’ve seen you present part of this at ASA last year in Philly, on your methods and it was really fascinating. So I wanted to ask about that too. You mention in the book that you ultimately worked as the Director of Customer Support at AllDone, the startup that you studied. How did that position within the organization help or hinder your fieldwork? What were some of the ethical dilemmas that you faced when managing workers that you were writing about?

Ben: There were two main advantages to how the research evolved, which echo Josh Seim’s formulation of the “observant participation” undertaken by deeply embedded fieldworkers. One was that the depth and breadth of my observations increased. Getting more involved in organizational processes allowed me to see and understand them in greater detail than I had before. And as a middle manager, the breadth of my vision also increased. I was overseeing customer support teams in the Philippines and Las Vegas while also serving as an information broker between San Francisco headquarters and those remote teams. So I got to see how each part of the firm fit together and how they all responded differently to the same organizational pressures. If I had maintained my role as an unpaid intern who was marginal to the organization, I never would have been able to develop the insights that came from my position in middle management.

The other advantage is tacit knowledge. I wasn’t sharing the exact same experiences as the people I was working alongside. But some aspects of my job helped me develop a better understanding of what they were facing. For example, I was the final authority when it came to customer support emails. Anything that couldn’t be handled by members of the remote teams would end up on my desk. Dealing with e-mails from angry customers gave me a taste of how hard it must have been for AllDone’s phone support workers in Las Vegas, who had to handle difficult users in real time. As an employee, I also had a tiny stock option grant. Even though I was a critical sociologist holding myself at an analytical remove from my participant role, I sometimes found myself wrapped up in the excitement of the company’s growth. When my colleagues imagined how their lives might change if they cashed out and got rich, sometimes I couldn’t help but fantasize alongside them. If I could be drawn into those imagined futures, I had no doubt about how meaningful they were for many of my colleagues.

There were also downsides to how my position in the organization evolved. Because this ended up becoming an ethnography of the managerial eye, my view of what workers in the Philippines and Las Vegas were saying or doing when managers weren’t around was limited. And of course there were myriad ethical issues to grapple with as well. I tried to inhabit a position of relative power in the organization as responsibly as possible. I advocated for better pay and working conditions for the low-wage workers who were beneath me in the organizational hierarchy—though most of the time executives didn’t listen to me. I was most successful at making changes that I could implement on my own, like giving workers advance notice of modifications to the platform that might affect their work. But there’s no getting around the fact that as much as I might try to improve things for AllDone’s remote workforce, by dint of my role in the field I was inevitably going to be participating in their domination and in the extraction of value from their efforts. I tried to follow the example of others who have found themselves in positions of power in the field by staying reflexive and continually analyzing my place in the organization and how my participation might be contributing to the reproduction of inequalities within the firm.  

Larry: Your analysis of start ups centers on how entrepreneurs juggle and manage multiple trajectories of growth—of valuation lags and organizational drags. What is specific about venture capital in shaping these trajectories for startups? And more generally, how should sociologists of science and technology study the role of capital in shaping scientific and technological work?

Ben: I come from a perspective that combines the sociology of technology with the sociology of work and economic sociology. I’ve been interested in the relationship between financialization and work for a long time. We have a lot of important research on how the rise of shareholder value ideologies and practices has transformed work and employment in publicly traded corporations. We need more research on other asset forms and how investors’ interests are related to workers’ experiences.

Venture capital is just one form of private equity financing; different asset forms have different logics. VCs pool money from other investors, throw in a bit of their own money as well, and then invest in a portfolio of startups. They expect that 9 out of every 10 startups they fund will fail. Either the companies will die, or their valuation won’t increase enough to deliver meaningful returns to investors.

The VC investment model has a corresponding managerial paradigm—which I call venture capitalism—that’s aimed at generating rapid growth. Each startup in a VC fund’s portfolio is trying to become that one in ten that’s a big hit. In platform startups like AllDone, software engineers are constantly experimenting with product features to find new ways to get key metrics up, like revenue and unique users. Investors look for startups that seem to be capable of growing fast enough that they’ll ideally be able to sell their stake to someone else down the road for more than they originally paid. One big success can more than make up for a portfolio littered with losses. For example, Sequoia Capital’s first investment in Airbnb was $585,000. That initial outlay turned into over $4 billion after Airbnb’s IPO—a 7,000-fold increase. Treating companies like assets is of course par for the course for institutional investors. But I argue that the VC model representsa uniquely speculative, supercharged version of financialized capitalism.

So what does this mean for sociologists of science and technology? I think it’s not enough for us to look at a company and say, “the design of this technology is influenced by the profit motive.” Of course, in one sense that will be true of any capitalist firm. But in the book I argue that we need to ask how is profit generated, and for whom? Is this a publicly traded company that’s beholden to Wall Street investors? Is this company owned by a private equity firm that acquired it through a leveraged buyout? Is the firm venture-backed? Looking at the structures of capital that animate a firm can help us explain what investors want, how they go about meeting their particular needs, and what pressures this places on an organization as it engages in technology development.

Larry: A part of the book that really fascinated me was the role of trust in getting workers to choose to provide their labor and services to the platform. You describe the different types of relational work (p. 139) that customer service contractors used to get workers to accept experimental changes on the platform. Could you describe some of the strategies and tactics of trust and relational work that is used?

Ben: I just talked to a reporter today who asked me whether I thought customer support agents were going to be automated out of existence! I don’t share that perspective. Tech companies often propagate the fantasy that a fully automated experience is just around the corner. What I saw at AllDone suggests why this isn’t the case. AllDone catered to independent contractors who provided local services to customers. Their livelihoods were at stake each time they used the platform. They needed to understand how to use it successfully. But the platform wasn’t designed to explain all the nuances and complexities of using the software—it was designed to bring in as many users as possible so AllDone would look good to VC investors. So education is one of the important functions of the relational work that helped to build trust with platform users. The service providers who used the platform to find work weren’t people weren’t always tech savvy. A lot of them were older. They might be a good electrician or house painter, but they weren’t so great at setting up a profile that would look attractive to customers.

There’s also repair work. Venture-backed platforms are constantly changing as developers experiment with new features to get the numbers up. Users can get upset when suddenly the platform isn’t working how it used to work. So how can a startup preserve relationships with those valuable customers? Running them through an automated system isn’t going to help. It takes real interpersonal interaction to help them feel heard, seen, and understood. Having a person-to-person conversation that lasts 30 to 45 minutes can go a long way in helping to establish trust in a technological system. This is an example of the behind-the-scenes labor that supports smooth user experiences with technology—what science and technology scholars sometimes call articulation work. Customer support workers at AllDone helped to keep the socio-technical system bound together when users’ trust was undermined or betrayed.

Larry: This reminds me of how during the pandemic, a lot of restaurants were automatically enrolled onto these like delivery platforms like DoorDash and UberEats. But they never knew that they were listed on these platforms and they got very pissed off at drivers coming in to pick up orders without warning. Other companies, in contrast, would go door to door and talk to the restaurant managers and try to convince them that their platform is good. That seems like a similar type of relational work that’s done to gain trust on these platforms.

Ben: Yes! I just saw a new paper by Anne Jonas that picked up on this idea. She studied virtual schooling in K-12 education. She also found that teachers performed a lot of relational work to try to keep online learners engaged. Platforms and screens don’t automatically educate students. As Allison Pugh reminds us in her new book, human connection is just as important as ever.

Larry: Your book also tells us quite a bit about how inequality is shaped by technology and platform work. One of the common refrains in the book is that: “Middle Aged women are what makes AllDone work”. But a lot of the pains and brunt of the changes that are made to these platforms are borne by them, while the riches and profits are given to other folks in different offices. What does this tell us about who stands to benefit from technological innovation, and who bares the brunt of the pains?

Ben: At AllDone, the pressure to move fast and break things was experienced differently depending on a worker’s structural position in the firm. For the workers in the San Francisco office who were orchestrating change, it was fun, exciting, and engrossing to be constantly testing out new ideas and discovering which ones would get the numbers up—think of Burawoy’s work games, which combine skill and chance with uncertain outcomes and intermittent rewards. For employees who held stock options, every step they took to bring in more users and revenue meant a higher valuation for the company and a higher possibility of cashing out.

Phone support workers in Las Vegas had a totally different experience of the same phenomenon. Software engineers saw users as abstract numbers on a spreadsheet. But frontline workers had to confront users as real people who had real emotional responses to being treated like a data point that could be experimented on and manipulated. Customer support agents struggled to keep up with the ins and outs of an ever-changing platform. At the same time, the users who were being directly affected by these changes were often quite upset, and they took their anger out on AllDone’s frontline workers. So standing between software developers and their users exposed phone support agents to a lot of stress, anxiety, and fear.

The team in the Philippines is interesting because it runs contrary to some common ideas about what we’d find in an offshored data-processing team. AllDone wasn’t exactly the “digital sweatshop” that some people imagine. Again, the software engineers were constantly looking for ways to change the platform to increase key metrics. So they delegated a lot of long-term organizational functions to information-processing workers in the Philippines. Instead of coding everything up, they could essentially task data workers with acting like algorithms themselves. The work was monotonous, but compared to other common options for college-educated Filipino workers (see Jeff Sallaz’s fantastic book on call center work), it wasn’t necessarily stressful. AllDone’s managers prioritized rapid growth above all else. They didn’t have much time to worry about efficiency, so it didn’t make sense for them to focus on trying to squeeze a little extra productivity out of workers who were being paid two dollars an hour.

If workers’ experiences on the job varied dramatically depending on their structural position inside the firm, so did their compensation. Ninety-two percent the company’s workforce were remote, work-from-home contractors. The majority of them were women. The managers in San Francisco often said that AllDone never would have achieved its incredible growth trajectory across multiple funding rounds without the remote teams. But the workers in the Philippines and Las Vegas weren’t technically AllDone employees, even though some had worked for the company for years. And because they didn’t hold stock options, they weren’t able to share in the incredible wealth they were generating. This is one of the ways that startups reproduce inequalities. In an economy where wealth is increasingly generated from the ownership of assets rather than from employment income, excluding such a large percentage of a startup’s workforce from asset ownership also means excluding them from the gains a tech company creates.

Larry: This probably like goes back to the alternate models of growth that you mentioned previously, on how there are alternative models of financing technological change. That’s something that is mentioned it in the conclusion of the book too. It’s certainly very interesting. Public options or state financed options of certain types of platforms potentially could make sense. What are your thoughts on alternative models?

Ben: When it comes to alternative financing models, we need to think about ownership structures that are designed to benefit a broad array of stakeholders, rather than just funneling wealth up to the folks at the top of the pyramid. One example is private ownership—it’s not a radical idea, but it does prevent outside investors from exercising control over how technologies are developed and how the gains will be distributed. Jessa Lingel, a colleague of mine at Penn, wrote a wonderful book about Craigslist, which was founded before most of today’s tech behemoths existed. Because it’s privately owned, it’s been able to follow an ethos that balances the profit motive with the public interest. Craigslist doesn’t harvest user data to sell it to data brokers. They only charge employers for posting job ads and landlords for posting apartment listings. The founder, Craig Newmark, has preserved the Web 1.0 ethos of an open internet where information is available to all. Craigslist is a profitable enterprise, but it’s never been a profit-maximizing enterprise. Avoiding external funders who demand rapid growth as a condition of their investment makes that possible.

The are also successful tech non-profits that, again, don’t face the expectation that they’ll maximize returns for investors. The example I give in the book, which Mary Gray and Siddharth Suri discuss in Ghost Work, is Amara. Amara is a digital labor platform that can be used for transcription projects. Wages are higher than on most labor platforms because the company isn’t trying to expand the bottom line. There’s also a lot of new research on platform cooperatives. These are digital platforms that are owned and operated by the workers who rely on them to find work. The example I give in the book is a platform called Up & Go, which is a house-cleaning platform for workers in New York City. Up & Go returns 95% of the income generated from house-cleaning jobs to the workers themselves. Instead of investing in scale and trying to dominate the market for housecleaners around the world so investors can cash out for billions, they’re just focused on helping local immigrant women increase their income and feel a sense of ownership over their work.

We should be thinking about what kinds of policies we can put in place to promote these kinds of alternative models for building tech companies. The VC model didn’t start booming until the 1980s, after policy changes loosened regulations that had kept pension fund managers from investing in VC funds. Cuts in the capital gains tax rate have also incentivized investments in venture-backed startups because they let investors hold onto more of their profits. If we want to loosen VC’s grip on innovation, we should reconsider these kinds of policies. We can also eliminate the carried interest loophole, which lets investors treat their investment gains as capital gains rather than as income. There’s also a qualified small business stock exemption that allows stock acquired in the early stages of a startup’s growth to escape taxation. Reducing the influence of VC investors could help to promote an environment where alternative ownership structures have more of an opportunity to flourish.

Larry: What are you working on next? Anything that you would like to preview to the SKAT newsletter audience?

Ben: I have a new project that’s still in its early stages, so I’ll tell you about three articles related to the book that are coming out soon. The first uses the relational work framework to understand the ties between AllDone’s software engineers in San Francisco and data workers in the Philippines. Managers in San Francisco sometimes felt guilty about the massive gulf in compensation and authority that separated them from Filipino workers. I show how workers in San Francisco alleviated some of their shame by framing the employment relationship to emphasize fun, friendship, and conviviality. Workers in the Philippines, for their part, really valued these jobs—they were flawed in some ways, but compared to other options, they could be quite attractive. I show the relational work they performed to make software engineers feel better about themselves, framing them as heroic job creators helping people in need. I argue that this relational work helped to maintain the stability of AllDone’s labor arrangements in a way that benefited both workers and the firm in the short term, while also reproducing the global inequalities that pervade the tech industry.

Another forthcoming article contributes to the nascent interdisciplinary literature on asset manager capitalism. We have some great research on how VCs exercise their structural power over firms, but less on how labor is organized in startups to achieve investors’ goals. I use the case of AllDone to illuminate the different forms of labor the company mobilized to inflate the firm’s value.

The third article, which I’m currently revising, is about my methodology. I consider what other ethnographers can learn from the dilemmas that I faced after unintentionally stumbling into the role of a middle manager while conducting fieldwork. The paper considers whether and how researchers can inhabit positions of power in the field in a way that is both ethical and generative of useful insights.