Jevons Paradox in the Age of AI
The curious case of the software developers who did not disappear
Like a lot of people concerned with the development of software, I find myself terrified and excited at the same time. Half the time I think I’m finished, that there’s going to be no more work for people like me. The other half the time I’m thinking, “Wow, I can do anything!”. The onslaught of new AI coding tools is relentless. Every day brings something new. Every day I have to think to myself: Is this important? Is it useful? Or should I just ignore it?
Tweets like this don’t help matters. That one in particular almost made me choke on my cornflakes this morning. But the reality is that yes, pure software engineering as a profession might disappear, but it’s likely to be replaced by something else. Take for example this much more positive blog post from Stack Overflow:
Why demand for code is infinite: How AI creates more developer jobs - Stack Overflow
stackoverflow.blog
I keep reading about something called “Jevons paradox” in the context of software engineering.
The original theory comes from an English economist named William Stanley Jevons. In 1865, he published a book called “The Coal Question”. At the time, Britain was in the middle of the Industrial Revolution, which was entirely fuelled by coal.
People were starting to worry that the country would run out of coal. Many prominent thinkers of the day believed that inventing more efficient machines would solve the problem. They pointed to the fact that James Watt had recently introduced a steam engine that used a fraction of the coal compared to the older Thomas Newcomen engines. The common belief was that this incredible efficiency would save the nation’s coal reserves.
The flaw in the logic
Jevons saw a major flaw in this thinking. He realised that making the steam engine more efficient did not just save coal. It made the energy produced by that coal much cheaper and far more profitable.
Before Watt’s engine, steam power was mostly used for one thing: pumping water out of coal mines. It was too expensive to use for much else. But when the efficient engine made steam power cheap, it suddenly made financial sense to use it everywhere. More machines, more factories, vastly more coal burned than before. The efficiency gains didn’t save coal at all. They just made coal useful for things nobody had considered.
Right now, AI coding assistants are the new steam engine. They are making software dramatically cheaper to produce. A developer can now do in an afternoon what used to take a week. The immediate fear is obvious: if a developer is twice as fast, a company only needs half as many. But Jevons paradox says that’s wrong.
Every company has a massive backlog of things they want to build but can’t, because developers are expensive and there aren’t enough hours. When the cost of building software drops, companies won’t just do the same work with fewer people. They’ll start building all the stuff they couldn’t justify before. Custom tools for teams of five. Software in industries like agriculture and local manufacturing that have never been able to afford it. That Stack Overflow article makes the point that demand for code is practically infinite, and I think that’s right. For every problem we solve, we find five more.
So what changes? The job changes. Instead of spending hours writing code, developers become the people who figure out what to build and whether it actually works. AI is good at generating small pieces of code. Humans are needed to understand what a business actually needs, to check for security problems, to navigate the politics of competing priorities, to decide whether a feature is worth shipping at all. Most importantly someone needs to keep an eye on quality. Bad code just causes problems down the line. That stuff doesn’t go away. If anything, there’s more of it when the volume of software goes up.
I think we’re heading somewhere interesting. “Developer” is going to mean something different in a few years. More of the job will be figuring out what to build and less of it will be writing the code yourself. And because building gets faster, there will be a lot more software in the world that needs someone paying attention to it. Whether that’s a good thing, I keep changing my mind. But probably, yes.
