Twelve years ago, while I was in computer programming school, I was engaged in a debate with a not-so-knowledgeable person about the value of a career in software development. He claimed that before too long the profession would be obsolete, as computer programs would be writing computer programs, and the humans wouldn’t be necessary. (He was a medical magnet salesman; I think he was trying to convince me that his profession had more merits.)

Thankfully, I didn’t argue the point much. There would have been little point in trying to convince him anyway, but at the time I didn’t really have a strong argument against it. Circa 1996, RAD and expert systems were big buzzwords, and so it seemed at least possible that, given enough time, research, AI, and computing power, programs-to-build-programs could become reality.

A decade and a couple of computing revolutions later, this dream is looking less likely. We now know a lot more about how little we know about software development. We know how complex and chaotic it is, and we know how poorly algorithms are suited to handling those sorts of problems. We know that the best work is done in organizations that emphasize people over process.

That’s not to say that computers, algorithms and AI haven’t made some appreciable advancements in capability over the past few years. Google is the best-known example. The Roomba and it’s big-brothers SWORDS and Predator have finally brought robotic sci-fi closer to reality. There’s real progress being made on the DARPA challenges. Can software-that-writes-software be that far behind?

I’m not going to attach a timeline to that question, but I do have one insight. It came to me when listening to Stack Overflow Podcast 21 (you can read the transcript here). Jeff and Joel were discussing the role of experts in particular fields, how deep their knowledge really goes, and how applicable it is to new development. Jeff said:

…the way that I’ve characterised [software development] in the past, it’s kind of like a number in binary. Right? So you start off with a bit… which is, you know, zero or one. And then as you go from left to right, you flip all those bits, and say you have six bits. By the time you’ve flipped six unique bits, think how many combinations there are of what you’re doing. It was always amazing to me as a developer, it didn’t take very long at all to get completely off the beaten path, to where, like, virtually nobody is doing what you are doing. And it’s not like you’re doing anything weird. It’s just that you made 6 unique decisions that are all independent of each other, and software development is just kind of like that.

6 trivial changes can make 64 different possible outcomes (2 to the power of 6). What Jeff means is that no situation in software development is exactly the same as any other; the number of variables is so large that the number of outcomes is enormous.

Computers today cannot deal with that sort of complexity. They won’t be able to do so in the immediate future either. With increases in algorithm research and raw power, they might be able to do so someday. I don’t know whether or not that will happen in my lifetime.

What I do know is this: software development is probably the most complex domain in existence. That means that, if computer programs someday replace the software developer, it will be one of the last careers to be replaced. Most of the others will have long since fallen to ever-more-capable robots and algorithms. If you want something that’s future-proof, software development is probably as close as you will ever get.