The Jevons Paradox of AI: Why Efficiency Won't Save Us From Ourselves
The more AI can do, the more we’ll ask it to do.
There’s a 160-year-old economics concept called Jevons Paradox that predicted this perfectly.
In 1865, an economist named William Jevons noticed something weird. Steam engines were getting more efficient. So coal consumption should drop, right? Less waste, less fuel needed.
The opposite happened. Better engines made steam power economically viable for way more industries. Total coal demand exploded.
Efficiency didn’t reduce consumption. It unlocked new consumption.
AI is doing the same thing right now.
The multiplication effect
Every time a task gets automated, it doesn’t disappear from the workload. It gets multiplied. Emails get easier to write, so we send more of them. Code gets faster to produce, so we build more software. An analysis that took a week now takes an hour… so we run 10 analyses instead of one.
The bottleneck moves, but it doesn’t disappear.
I see this in my own work constantly. I used to spend days building a feature. Now I can prototype it in an afternoon. But I haven’t gained days of free time. I’ve filled that space with more features, more experiments, more products. The effort collapsed but the ambition expanded to fill the gap. And then some.
This isn’t a failure of discipline. It’s the natural response to a constraint being removed. When something becomes cheap, you use more of it. Always. Jevons saw it with coal. We’re living it with cognition.
The inbox problem
Think about email for a second. AI can now draft a response in seconds. Fantastic. But if everyone has that capability, the volume of email goes up, not down. Every inbox gets fuller. Every message gets a faster reply, which generates another faster reply, which generates another one.
The tool that was supposed to save you time just created a faster treadmill.
This pattern is showing up everywhere. AI-generated marketing means more content competing for the same attention. AI-assisted code means more software competing for the same users. AI-powered analysis means more reports competing for the same decision-maker’s time.
The supply side is exploding. The demand side hasn’t moved.
Where the bottleneck lands
If the production cost of almost everything approaches zero, what becomes scarce?
Attention. Judgment. Taste.
The ability to look at ten AI-generated analyses and know which three actually matter. The ability to ship five products and recognise which one deserves your full commitment. The ability to receive fifty AI-drafted emails and decide which five are worth a thoughtful response.
We’re not heading toward a world with less demand for human thinking. We’re heading toward one that demands more of it, just in a different shape. Less “can you build this?” More “should you build this?” Less execution. More curation.
The people who thrive won’t be the ones who can produce the most. They’ll be the ones who can filter the most. Who can look at a wall of AI-generated options and say “that one” with conviction.
The organisational version
This plays out at company scale too. I’ve watched organisations adopt AI tools expecting to do the same work with fewer people. That’s not what happens. They end up doing more work with the same people. Scope expands. Expectations inflate. The CEO sees a report that used to take two weeks get delivered in two days and thinks “great, now do five of them.”
Nobody planned for the demand to increase. But it always does.
And this creates a strange new kind of burnout. Not from the work being hard, but from the work being endless. When there’s no natural friction slowing things down, there’s no natural stopping point either. The steam engine doesn’t get tired. Neither does the AI. But the people directing them do.
The upside nobody talks about
It’s not all bleak though. Jevons Paradox has a flip side that often gets overlooked.
Yes, coal consumption exploded. But so did economic prosperity. The industrial revolution happened because of that explosion in demand. More consumption meant more industry, more jobs, more wealth, more progress. The paradox wasn’t a trap. It was a catalyst.
The same thing is probably true for AI. More software being built means more problems being solved. More analyses being run means better decisions being made (on average, at least). More experiments being launched means faster discovery of what actually works.
The magpie in me sees this clearly. The collapse of development costs doesn’t just mean more stuff gets built. It means more ideas get tested. And that’s how you find the ones worth pursuing.
The paradox isn’t that efficiency creates more work. It’s that more work creates more opportunity. The question is whether you can see clearly enough through the noise to grab it.
So what do you actually do about it?
You can’t fight it. The answer isn’t to artificially limit how much you produce or to pretend the efficiency gains don’t exist.
The answer is to get really good at the thing that can’t be automated yet: knowing what’s worth doing.
Build your judgment faster than AI builds your output. Develop taste. Learn to kill your darlings quickly. Get comfortable saying “just because I can doesn’t mean I should.”
Because the steam engines are only getting better. And the coal pile isn’t going to manage itself.
Enjoyed this post? Leave a comment or like it on Substack.