Programming & Jevons Paradox

Over the last few years, large language model providers have spent a considerable amount of time, effort and marketing budget attempting to convince us that programming is dead. We have been told, repeatedly, that software engineers are only months away from redundancy and that increasingly capable models, wrapped in the right tooling, will write all the code.

The more extreme versions of these predictions are not entirely unfounded. AI has already changed the economics of software development. The mistake is assuming that because code becomes cheaper to produce, the demand for programmers must therefore collapse.

We've heard this story before.

Many executives have been quick to announce layoffs under the banner of artificial intelligence. Sometimes that may be justified. Often it appears to be a convenient explanation for over-hiring, poor strategic decisions, missed forecasts or changing market conditions. Blaming a technological shift is usually easier than admitting mistakes. To be fair, some leaders acknowledge this openly. Many do not.

What's interesting is that inside companies the story often sounds very different. Rather than reducing the amount of work, AI is frequently increasing it. Teams that can produce software faster are discovering more opportunities to build, automate and experiment. Features that were previously uneconomical suddenly become viable. Internal tools get built. Customer requests that once sat in a backlog for months or even years are becoming do-able.

This is not a new phenomenon. Economists have observed it repeatedly throughout history. Jevons Paradox describes a phenomenon whereby increasing efficiency leads to an increase in overall consumption rather than a decrease. Put simply, when something becomes dramatically cheaper, people often end up using more of it.

Of course, this feels completely backwards. Most people intuitively assume that if a resource becomes more efficient, we will need less of it. Yet history is littered with examples showing the opposite. Lower costs create new opportunities, new markets and new behaviours. Efficiency doesn't always reduce demand. Sometimes it explodes it.

The question is whether software follows the same pattern. If code generation becomes effectively free, organisations may not respond by employing dramatically fewer engineers. They may respond by building dramatically more software.

Historically, when something becomes cheaper, people rarely consume less of it. They consume more. The bottleneck then moves elsewhere. In software, that bottleneck is unlikely to be code. It is more likely to be judgement, coordination, product insight and the ability to distinguish a good idea from a possible one.

The irony is that "AI will write all the code" may turn out to be substantially true while "therefore we need no programmers" turns out to be substantially false.