If software gets cheaper to build, you might expect businesses to spend less on it. Some will. Others will finally build the things they’ve been putting off.
That second possibility is what interests me about Jevons’ paradox. A lower cost can change how much of something people want, and what they can practically do with it.
In his 1865 book, The Coal Question, William Stanley Jevons argued that using coal more efficiently could increase total coal consumption. More efficient engines made power cheaper and useful in more places. The growth in use could outweigh the fuel saved by each engine. Read Jevons’ original explanation.
The broader idea is called a rebound effect: efficiency lowers the cost of doing something, so people do more of it. Jevons’ paradox is the stronger case, where that extra demand more than offsets the resource savings. The German Environment Agency’s explanation of rebound effects gives a straightforward introduction to how lower costs can change behaviour.
It is a possible outcome, rather than a rule that efficiency always increases consumption. The size of the response matters.
Software isn’t coal. Applying this idea to development budgets is an analogy, rather than proof that software spending will rise.
But the mechanism is easy to recognise. Plenty of business problems stay unsolved because the software would cost more than the problem justifies. A useful internal tool might save hours each week but still be too expensive to build. An integration might remain a manual process for the same reason.
If development becomes cheaper, some of those decisions change. More projects become worth doing. Smaller businesses may be able to commission software they previously couldn’t afford.
Imagine a business with one project expected to cost $100,000. Suppose, purely for illustration, the same scope could now be delivered for $50,000.
It could build that project and keep the other $50,000. It could spend the original budget on that project plus another useful tool. Or it could invest more overall because several previously unaffordable projects now make financial sense.
Those numbers are made up to show the choices, not an estimate of AI savings. In each case, the first project became cheaper. Total spending depends on what the business chooses to do next.
More software being built doesn’t itself prove Jevons’ paradox, either. The original concept concerns total resource use. Software output, development spending, engineering hours and computing demand are different things.
When I hear that AI will make software cheaper, I don’t assume the result is simply smaller development budgets. I also think about the backlog of things businesses would build if the cost made sense.
Some will take the saving. Some will do more with the same money. Some will spend more because there are now more worthwhile projects. Which happens will depend on demand, budgets and whether the software actually solves a useful problem.
Cheaper development can expand what gets built. It doesn’t tell us, on its own, whether total spending goes up or down.
For the evidence on development productivity itself, read How much should software cost now with agentic AI?.
