The limited value of a computer science education

I started programming at the age of 10 on my TI-82 graphing calculator. I loved video games as a kid and was ecstatic to learn I could make them myself – while also distracting myself from the boredom of math class. Very rapidly I came to love the craft of programming in and of itself and entered into a long and rewarding journey of nurturing my skills as a programmer.

Part of that journey included a bachelor's and master's degree in computer science at Stanford, including a significant amount of time as part of a research group. As a professional I've built many useful projects and contributed significant knowledge to my field. In this post I will explore what value, if any, a computer science education provides for a professional developer.

There are people who take the extreme positions, that a computer science education is totally useless or that a computer science education is completely essential. My position is that a computer science education is overvalued, though not useless. The vast majority of what a programmer does, forming good abstractions and avoiding complexity, is almost completely untouched by computer science curriculums.

A handful of times in my career, I encountered problems which I don't think I would have solved without my computer science education. These all related to algorithms in distributed systems that required formal proofs to be confident of their correctness. These few cases happened to be critical to their respective systems – for example, this algorithm is what made Storm possible. So in that sense, I personally have gotten a lot of benefit from my computer science education.

However, I also have to acknowledge that the scale and difficulty of the problems I've worked on is unusual compared to the vast majority of programmers. Most people aren't designing new distributed processing paradigms. I highly doubt most programmers need to know how to do formal proofs of algorithms, determine asymptotic complexity of algorithms, or even know that much about data structures. Most people are doing the kind of programming that involves piecing together components like databases, application frameworks, deployment tools, monitoring systems, and data processing frameworks into an application.

Of course there are some fields for which having a computer science education is much more relevant on a daily basis – distributed systems is one that comes to mind. But even in that field, the vast majority of your work is forming good abstractions and avoiding complexity – those two crucial skills that you don't learn in a computer science education. The benefits of a computer science education are limited to the value you get out of a small percentage of situations. For me those situations were very high value, but most programmers don't even run into them.

The reason why I even feel the need to write about this topic is because computer science skills are generally all that is tested in programmer interviews, via the infamous whiteboard coding interview. Whether someone can or cannot solve some cute algorithm problem in a high-pressure situation tells you nothing about that person's ability to write solid, clean, well-structured programs in normal working conditions. This practice is akin to hiring pilots based on how they do on an aeronautical engineering exam. That knowledge is relevant and useful but tells you little about the skill that person has for the job.

What's especially bizarre about this practice of recruiting is there exists an alternative that is better in every respect. Take-home projects are a decent measure of programming ability and are far less of a time investment for the company (15 minutes to evaluate project vs. 1 hour for interview). Plus I find that most candidates prefer take-home projects because it's less stressful and gives a better opportunity to show off their skills.

So in that sense, due to irrationality in the software industry a computer science education is useful for doing well in job interviews. But there have been some trends recently indicating that these nonsensical recruiting practices will change – programs such as Recurse Center and Insight Data Engineering help their residents get jobs by using the projects they've built over the course of one to three months as their selling point. So as the industry adapts and computer science knowledge ceases to be so crucial to getting a programmer job, a complete computer science education will become less important. How long that will take to happen is unclear.

In my experience, the only way to become a good programmer is to do it... a lot. Reading books and taking classes won't make you a better programmer – they might provide some guidance, but real improvement will come from the grueling hours you put into real projects. I could read books all day about how to execute a perfect barrel roll, but it will all be meaningless until I put on a parachute and get in the cockpit. Programming is the same way.

Overall I have mixed feelings about the value of a computer science education, mostly because of the personal benefit I have gotten from mine. For most cases though, I think it is severely overvalued. It's very strange to observe an industry with major talent shortages, and then to know perfectly good self-taught programmers get prematurely rejected in interviews because they don't have a computer science background. I hope to see the industry improve in this respect, but in the meantime I'm happy to exploit this imbalance as a competitive advantage.