About 2.5 years ago, I finished a Ph.D. with Sam Madden, David Karger, and Rob Miller, with my thesis at the intersection of Databases and Crowdsourcing. April marked my third year since joining Locu, which was acquired by GoDaddy about a year and a half ago. Having recently moved on from the company, I feel like this is a decent point for reflection.
Doing computer science research as a grad student can be a pretty amazing experience: you have several years to make cool things and scratch a bunch of mental itches. You get to think long and hard about problems, create new ones, and spend healthy amounts of time thinking about solutions.
Really early stage startups are also a place to make cool things, but tend to offer a frenetic whiplash-inducing experience as you try to find products that meaningfully improve the lives of some set of customers. Until you find something that fits, your job at a startup is to iterate quickly and favor fast failures over elegant contributions.
One commonality of grad school research and an early stage startup is the opportunity to work on and build cool and interesting things. Since they differ in their approach substantially, a natural question arises: Are startups a good place to go after you finish grad school?
Hopefully my experience arc of finishing a Ph.D., joining a startup, experiencing an acquisition, participating in an IPO, and deciding to move on can serve as some sort of lesson to others. It’s an N=1 guide and your experiences will be different, so caveat emptor and other warnings in dead languages. I’ve packaged the guide into an incomplete series of lessons targeted at grad students and other folks in the research world who might also be interested in exploring work at a startup.
Expect some funny reactions
I get the sense that by the time I left grad school, startups were a more common post-graduation choice than when I started school. My advisors, family, and friends were really supportive of my decision, but I sometimes got the sense that other people were surprised or taken aback by my decision. This was particularly poignant at conferences I attended a few months before and after defending my thesis. Although most folks took the news as they’d take any other exciting career choice, I also observed a decent amount of fidgeting and uncertainty in some people’s reactions.
It makes sense: the objectives you optimize for as a successful grad student are best suited for establishing a career in academia or industry research. As a result, a natural gut reaction to news of someone’s entering startups is that they are throwing away the work they did in grad school. In my research area, it’s more widely accepted that a natural conclusion of the systems one builds might be their commercialization, but this is less accepted in other research areas. At computer science departments that have had professors or alumni with successful startups, it’s easier to accept a post-grad school startup as a logical next step. What I’ve unscientifically heard from folks at departments with less examples of professors or alumni diving into startups is that there’s a stigma attached to it.
To be clear, entering a startup after grad school will likely reduce your chances of re-entering academia down the line, as you’re spending less time than your peers focusing on the sorts of things that academia values. This notion of narrowing down ones’ career choices will naturally result in discomfort or concern. Luckily, we’re increasingly seeing examples of people living at the intersection of both worlds, and I hope the stigma/false dichotomy fades with time.
Everyone’s story is different
As I neared the end of grad school, my advisors recommended that I wrap up my thesis and spend the following year conducting an academic job search. That sounded nice: I loved my time in academia, and still love teaching and mentoring people, so I planned on trying that out.
Then the opportunity at Locu opened up. Locu’s founders, who I met during our overlap in grad school, approached me to head up the company’s data and crowdsourcing efforts, which was broadly my area of research. In grad school, I did a lot of systems building around human processes, and I felt like I could do really interesting research in an environment where we could harness more than half a million hours of crowd worker contributions to clean and structure data.
Just like an applied particle physicist would probably benefit a lot from working at CERN, it makes sense that an applied social computing + data systems builder would benefit a lot from working at a place like Locu. If I were, for example, a theoretical computer scientist, that decision would be pretty different.
Managing a team that built systems in my research area made a lot of sense in my situation, but you should hardly listen to people who blindly peddle advice that startups are good for everyone coming out of grad school.
You can try before you buy
I finished up at MIT about 2.5 years ago, but started at Locu part time before that, about 3 years ago. The day after I submitted the last paper (chapter) of my thesis to a conference, I started spending around two days a week at Locu while compiling and editing the document that became my thesis.
This arrangement provided the benefit of trying out a startup before I bought into it fully. I got to learn two important things through it. Most importantly, I learned whether I worked well with the team, and whether the founders and I could work well together. I also got to see whether I enjoyed the type of work I’d be doing. As I describe later, things were initially too hectic to do research, so I had to take a bit of a leap of faith on the research opportunities that would pop up. Given the team’s early excitement at my doing other activities like blogging about what we did, I felt like they would be receptive to answering deeper questions as well.
The arrangement had the downside of being pretty stressful. Finishing up the thesis and putting together a defense was more time consuming than I planned. Startups are hectic places, and the combination of the two made for half a year of stressful times. I enjoyed both sets of activities, but my mind was never at rest, and I had a little scare when, after my defense, my thesis committee asked for a bit more work than I expected. I had to scramble to put in final edits while already working full time at Locu.
It worked out, and I’m grateful I got to try Locu out before I bought in completely. That said, there are likely saner ways to do this: consider interning at a startup for a summer rather than trying to multitask two complex sets of responsibilities.
You can do research at startups
Most people will advise you that it’s hard or impossible to do meaningful research at a startup. I agree that it’s hard, but it’s definitely not impossible. In fact, the kind of research I was able to do at Locu would have been challenging to work through in academia. Amazing crowdsourcing research comes out of academia all the time, but it’s rare that academics have established multi-year relationships with hundreds of paid crowd workers. This long-term, high-resource relationship had an impact on what I view as meaningful research in the area of practical systems to support crowd work.
That said, my job title and job descriptions at Locu or GoDaddy didn’t contain the word “Researcher.” My bosses were very supportive of introspection, though, and were happy if I spent some time on researchy questions. My day-to-day job, however, involved leading teams that made an impact on our customers and products, and when we were a young startup in particular, most of the sleep I lost was on things that weren’t research-related. For the first year and a half or so at Locu, it was hard to do what one might consider academic research.
As the company matured and our business processes became more sane, and in particular after we were acquired and were given the space to think more broadly, it became easier to do what looks more traditionally like research. We’ve got two full papers in submission on the work we’ve done, and it feels nice to be contributing back to the community I came from.
At peak research throughput, my weekly work schedule was still almost entirely not research-related. In general, I spent early mornings and weekends working on things like writing and making figures. In the weeks leading up to paper deadlines, I would spend about half of my working day implementing analyses. That said, as someone who wasn’t explicitly a researcher, I typically kept these activities outside of 9AM - 5 PM.
In short, if you’re building something interesting at a startup, it’s possible that with time you can ask and answer questions that are interesting to your research community. There are two hacks I’ve found for doing this effectively:
- Just like academic advisors hire grad students to augment their research efforts, you can do the same! Hire a graduate student as a summer intern: they get access to some amazing datasets, and you benefit from having an immensely talented person to collaborate with on hard problems for a summer. I was so lucky to work with Daniel Haas from Berkeley when he joined us for a summer as an intern. One of those in-submission papers I mentioned is thanks to the work he did with us one summer on automatically identifying crowd worker output that could stand to be reviewed by more experienced workers. In addition to saving us money and improving work quality, it turned out to be one of my favorite papers to write!
- Collaborate with folks that are still in academia. These are people who have the drive and external incentives to deliver meaningful research contributions, and at least in my field are quite interested in working with people embedded in industry. Aditya Parameswaran and I just submitted a first draft of a book on Crowdsourced Data Management to our editor, and I can’t imagine either of us would have been able to do it without the other. Plus, it’s a great excuse to hang out with some wonderfully cool folks like Aditya.
The community might be interested in your work even if you don’t publish
Even if you don’t publish papers on your work, the research community might still be quite interested in you and your research. Luckily, there are several ways to stay involved.
One way to stay involved is to participate in program committees, where you review paper submissions and discuss which papers should be presented at conferences. Greedily speaking, it’s a fun way to keep in touch with your research friends, but it’s also a nice way to stay in touch with what’s happening in the research world. Reviewing papers also serves as a mentorship opportunity: find some coworkers that are thinking about research/grad school and mentor them in how to review papers. They get to kick the tires on a new experience, and you get help reviewing your papers!
Another way to stay involved is to participate in industry panels or submit industry track papers to conferences. Most academic conferences want to learn what’s relevant to practitioners, and as someone that’s been in both worlds, you’re a perfect person to help bridge that gap! One slightly frustrating problem we’ve run into is that our papers have a harder time getting accepted to purely academic research tracks at conferences. But the reviewers always encourage us to submit to industry tracks, which seem a lot more welcoming to our line of work.
There’s also a whole world of non-academic conferences that I didn’t really participate in before graduating. You picked up all of those presentation skills in grad school, and these are nice opportunities to use them! O’Reilly runs a set of conferences like Strata. There are also a bunch of smaller but more intimate communities like Craft Conference, or !!con that might be nice venues for your work. Strata was fun, and while I haven’t been brave enough to submit proposals to the tighter-knit conferences, I hope to join those communities one day and you should too!
One word of warning: These conferences tend to be expensive to attend. By giving a talk, you generally don’t have to pay a registration fee, and the conferences will often fund travel/lodging, so make sure to ask!
A Ph.D. isn’t a perfect degree for startups
Your objectives and reward systems are different in academia and startups, and as a result, you have to hone a different set of skills.
In academia, you optimize for clean, elegant, novel solutions to broad problems that you evaluate in a deep way. In a Ph.D., you spend your time honing your question asking, answering, and presentation skills, and are critiqued on your clarity, generalizability, depth, and novelty. If you can present an interesting question, answer, and evaluation mostly through papers and slides, people are willing to look past code that’s poorly documented, untested, runs on only a single machine, and would likely need to be reimplemented if it was commercialized. (None of these comments are meant to come off as disparaging: researchers are simply optimizing for different objectives.)
In startups, you’re building an ecosystem around some core technological or societal insight. You’re not rewarded for novelty as much as you are for some combination of utility and revenue. Early on, this means rapidly iterating with customers and dropping solutions that don’t work for ones that do. People care less for why something works than for if it works. As you come upon a solution, you focus less on presenting it to your community and more on stabilizing it, automating it, scaling it, and reporting on it.
The human aspects are quite different as well. Whereas academia rewards the degree to which you establish your unique identity and build relationships across institutions, startups optimize for growing a product and a team. Both academia and startups could stand to improve how they think about team health, professional growth, and people’s sense of self-worth, but that’s for another day.
In academia, your initial time horizon for hitting something interesting is a bit longer than in startups, and as a result, you can afford to take the long road to your next experiment. If you discover a tangentially interesting thing along the way, the reward might be your next research project. In startup land, after an initial set of iterations and discovery, it’s your job to set up a process that will keep your company useful, growing, and relevant for years or decades, and only a small part of that involves coming up with something brand new or arguing for its novelty.
Several times at Locu, I found myself short circuiting a tangent that, toward the end of grad school I would have identified as the start of a new project. At the startup, we had to avoid these tangents because they would have significantly distracted from our core focus at the time. I didn’t always ignore the tangents though, and some resulted in internships that released open source or papers into the world.
If your academic interests and the focus of a startup align, that might help motivate you to deliver good things in startup land. But don’t confuse your interests for your skills. You’ll have to learn a new set of objectives and approaches if you transition from grad school to a startup.
A Ph.D. is a useful degree for startups
Perhaps if I had gotten six years of experience in industry rather than going to grad school, I’d be in the same position professionally that I am in now. For what it’s worth, I don’t think that I would be in the same position. At both Locu and GoDaddy, I used the skills I picked up in grad school to solve problems and collaborate with and mentor other people. I had to learn a ton beyond what I learned in grad school, but I’m grateful for the useful skills that the Ph.D. offered me.
To start, here’s a list of things you do in grad school, biased toward systems builders: find a good way to state problems, identify solutions to those problems, work on certain problems for months or years, architect and build systems, identify reasonable algorithms, measure things that are hard to measure, mentor some undergrads, collaborate with humans, and communicate clearly in written, spoken, and visual forms. Every single one of these skills is useful basically anywhere you will go, including startups.
At Locu, I mentored and managed some amazing fresh-out-of-undergrad computer scientists. It’s not an exaggeration to say that every single one of them was faster than me at solving well-stated problems. That said, they needed help thinking through problems and solutions, keeping the higher level objective in their heads, and getting comfortable with uncertainty. If there’s one thing grad school prepares you for well, it’s smacking your head and keyboard against an uncertain problem for several months on end until something meaningful falls out, all the while making sure you’re contributing to some cohesive story. A combination of teaching, mentoring, problem-solving, and presentation experiences I picked up in grad school helped me team up with these amazingly talented builders so that we could do something nice.
The other area grad school really helps with is in external communication. At Locu, our founders were certainly the most active communicators. After them, it seems like the more academic members of the team tended to put ourselves out there the most, both in terms of offering to make presentations and in finding paper- or blog post-writing opportunities. These are skills you get comfortable with in academic life, and translate well to startups and beyond.
Don’t listen to any one person’s advice on the ideal life after grad school. Some of my friends from grad school love their lives as professors, and others quite enjoy their lives at larger corporations or industry labs. For me, startups provided a nice way to keep pursuing my grad school interests while working in a different context and at a different scale, and I’m grateful for the experience.
Many thanks to Aditya Parameswaran, Arvind Thiagarajan, Jean Yang, Lydia Gu, Meredith Blumenstock, Michele Catasta, Neha Narula, Nitesh Banta, and Philip Guo for reading a rough draft of this.