Is a Computer Science Degree Necessary?

Written February 9th, 2014, published February 15th, 2016. [1]

I get this question from time to time and I used to waffle on the answer. That, I think, is just a symptom of not framing the question clearly.

Let’s disentangle some questions:

1. Is the learning that happens in a CS program necessary?
2. Is the diploma necessary?
3. Are there things you learn in a CS program that are difficult to learn in the field?
4. Conversely, are there things you learn in the field that are absent in the CS program?

Is the learning that happens in a CS program necessary?

It depends on what sort of programming you want to do. The thing about software engineering is that the most common and visible of jobs programmers get is web development. Web development is like construction. The majority of things that are built are probably residential, and usually the difficult bits are outsourced anyway. The house builder doesn’t design and build a custom boiler, for instance. He gets something built by a specialist and plugs it into the house via a well defined set of interfaces. Similarly, the web developer is picking up the difficult pieces (an operating system, database, even web framework), and assembling it using a well defined set of interfaces. In fact, you’d probably get better at this sort of engineering working in the field rather than sitting in a classroom. [2]

But let’s crank it up a notch and say that you’re a homebuilder and you get a job to build a 50 floor office tower. You’d probably have to know more about physics, chemistry, mechanical engineering, not to mention building codes, project management, inspection requirements, and all the other meta stuff that goes into a substantial engineering project. That’s the analogy I’d like to make to having a CS degree. A web developer moving on to building a highly available distributed database is going to require both theory and practical knowledge not found in that 90% of development jobs. School is a finely tuned instrument for pushing theory to poor souls.

Is the diploma necessary?

That’s an easier one to answer. If you’re doing research, most likely. In that situation, it’s important to have a broad view of the field, because you can’t spend time inventing things other people already invented, and you have to build on results other people have discovered. And like I said earlier, school is a finely tuned instrument for pushing theory.

For everything else, you probably don’t need the diploma. In fact, startup land has gotten very good at ignoring it. Why is that?

First off, for most of software development, that 90%, a few years of experience far outweighs what you do in school. As an interview question, I used to ask candidates to estimate the amount of time they’d spent coding or thinking about code. What people do in their first year or two out of school quickly outpaces all of what they did their entire lives leading up to that point. Programming, for most people, is essentially a side hobby up until the first full time job. So, if you’re hiring for that 90%, you probably shouldn’t care about the degree.

Though it does tell you some useful things. If a candidate went to a top notch school, nepotism aside, you can probably infer that candidate has the capability to focus and get good grades, ie the ability to do what is expected. If the candidate did well in a top notch school, then you can probably infer they know many potentially useful CS topics, and were able to adapt to a higher level of expectations.

But insisting on a top notch school with top notch grades precludes some particularly large classes of valuable programmers – those who simply don’t care to appease others, are unconventional in their thought and beliefs, or focus so closely on computers that they fail in other areas. You probably want some of these folk on staff.

The days of “our startup consists of all CS PhDs!” being a selling point are long gone.

Are there things in a CS program that are difficult to learn in the field?

Absolutely. You can say you know a lot about an operating system, but unless you took an OS class in school, you probably have never actually implemented one. The experience of writing parts of an OS are key in truly understanding how it works and what you can expect from it in the field. It’s similar for programming languages. You don’t truly learn about your arguably most important tool without having to implement a compiler or interpreter.

Data structures that are not already baked directly into a language are difficult to learn about if you don’t already know of their existence. A good grasp of theory helps when looking for more esoteric techniques others may have already discovered.

Another thing that should not be underestimated is the amount of feedback you get while in school. The classes are designed to tell you if you’re wrong about something quickly. It’s very easy to find a job where you write code that nobody else sees. Not everybody does code reviews, especially really small startups. So there’s often times little feedback, and without a feedback loop, it’s difficult to get better.

Are there things you learn in the field that are absent in the CS program?

Absolutely. How to engineer software, for one. Classroom projects usually don’t live for more than a semester, and much of real world engineering has to do with maintaining systems that exist past the time at which any particular engineer stops working on it. I have never heard of a class that directly addresses this issue.

When I was in school, I signed up for a class with the unfortunately broad title of “Software Engineering.” The professor got up at the podium and actually said, “I’m not really sure what I’m supposed to teach you guys this semester.” I dropped the class.

This is what we should have learned:

• Coexisting peacefully with peers using a source control system.
• Communicating clearly through documentation, bug trackers, and wikis.
• Effective testing and automation.
• Development methodologies, project planning.
• Design patterns, techniques for structuring large systems from small components.
• How to work with a large code base written by people you have no access to.
• Using debugging tools and analyzing log files.

That last one is interesting. At a previous place I worked, we would give candidates an nginx log file and ask basic questions such as “what is the most common path?” or “what percentage of requests are returning 40x errors?” Newly minted graduates, even from prestigious CS programs, would almost always struggle with those questions. Candidates who mustered correct answers often did so by writing a Python or Ruby program using regex. Very few knew how to use grep, awk, and sort.

So, do I need a CS degree or what?

I still can’t definitely answer that question. But another way of looking at it is, what’s the downside? You learn a lot about how to approach problems – generally good for the brain – at the cost of a few years and some cash. You have a wider selection of programming jobs to pick from, most of which will probably pay more (that distributed systems specialist will command a higher salary than a run of the mill backend engineer), which lets you recoup that lost tuition money fairly quickly. It’s a pretty reasonable investment of money and time, assuming you can get a spot in a decent curriculum. But if you can’t stand boring lectures, or you don’t believe the rest of the college experience is worthwhile, by all means, skip the degree.

The harder question is, what constitutes a decent curriculum that prepares you for that 10%? I think one common characteristic of all of them is that they organize around a breadth of topics and don’t worry so much about the details. Berkeley, for instance, has no required class that teaches you how to program in a particular language. At the opposite end of the spectrum, I have yet to encounter someone with only a General Assembly / Skillshare / Codecademy / OpenCourseWare filled resume who is very good at programming. I’m not pessimistic about those curriculums or online learning yet. I simply have yet to see that pan out, so maybe that’s something to look out for in 2014.

It might be possible to squish the right curriculum into a year or so of intensive study. And this could be done, perhaps, General Assembly style, or online like Codecademy. It’s ambitious, but it doesn’t seem entirely outlandish to me, at least as I type this. I’ll close with what I think my dream curriculum would be, completed in four quarters of full time study.

1. Core data structures, everything implemented in the Python standard library for example.
2. Intro level algorithms, topics drawn from Knuth volumes 1 - 3.
3. Survey treatment of assorted topics from operating systems, programming languages, databases, networking, randomized algorithms, and machine learning.
4. “Software engineering” a la the topics I listed above, requiring building and shipping a large real world project. I imagine this would be a several month project involving a multi person group.

[1]	Yes, it took me over 2 years to publish this because I was trying to write the best static site generator. Those who can build their own tools are doomed to build their own tools. (I never finished it.)

[2]	MIT switched their famous 6.001 curriculum for this reason, so maybe it can be done in academia.