Beyond Jevons Paradox

Jevons Paradox is the most cited piece of economics in any conversation about AI. Make a resource cheaper and we use more of it, not less. Coal is the classic case: steam engines got more efficient and coal consumption went up, because lower cost opened up uses that were never worth it before.

Apply that to AI and you get the obvious conclusion. Cheaper, faster, more capable models mean far more AI gets used. That's almost certainly right, and it's also the shallow half of the story.

Jevons explains why cheaper AI leads to more AI. It says nothing about which ideas people decide are worth pursuing in the first place.

The deeper change isn't that we can do more. It's that more ideas survive long enough to get attempted.

The brain isn't only a thinking machine. It's a rationing machine. Every day it decides where to spend time, energy, attention, money, knowledge and effort, and all of those run out.

So it can't seriously evaluate everything available to you. The world offers far more than anyone could pursue — things to learn, build, fix, automate, improve, explore. Consciously weighing each one would burn all the attention you have.

Instead the brain kills most options before they ever reach a conscious plan, by guessing what they cost.

Not financial cost. Total personal cost: time, effort, energy, knowledge, risk, complexity, what you give up to do it, and the damage it does to everything else in your life.

So when an idea shows up — build an application, write a book, start a company, learn a language — the first question is almost never whether it's interesting. It's whether it's doable.

That happens in about a second, mostly below awareness. How long would this take? What would I have to learn? How many evenings disappear? What do I stop doing? Does it wreck work, or family, or the rest of the queue?

Too expensive, and the idea gets discarded. Not because it's bad, and not because you don't want it. Because it doesn't fit the life you're actually living.

Do that for enough years and it hardens into an internal cost filter.

The filter stops you wasting attention on things that look impossible or unaffordable. That's not irrational, it's necessary. Without it you'd be permanently distracted by possibilities you could never act on.

It's also trained by experience. Take someone with a full-time job, a family and a few free hours a week. They have ideas — a résumé service, a proposal generator, a stock-analysis tool, a personal automation setup, a budgeting app, a language tutor.

None of those are impossible. They're unrealistic given what's available.

The brain learns that lesson and stops running the calculation. Next time something similar appears it gets discarded almost instantly, not for lack of value, but because experience already predicted the outcome.

So most ideas die before you notice you rejected them. They never become plans or experiments. They're a vague fantasy for a moment, and then they're gone.

AI does more than speed up execution. It changes the number the filter is working from.

A project that took six months becomes a weekend prototype. A task that needed a team becomes realistic for one person. Research, planning, coding, design, documentation, testing and iteration all compress at once.

None of those savings is revolutionary alone. Together they move the estimate.

Every hour taken out of execution changes what the brain thinks is achievable. Enough small reductions and whole categories of project cross the line from not worth attempting to worth trying.

The idea didn't change. The price did. And when the price changes, the filter changes its answer.

Ideas that used to trigger an instant no start getting a maybe, and some of those turn into yes. That isn't a productivity improvement. It's a change in how much future a person believes they have available.

Here's the part that gets missed: AI mostly isn't generating the ideas. They already existed.

Someone wanted to build a particular service for years. They understood the problem, saw the opening, could picture the product — and knew it would eat months of evenings. So it got filed under unrealistic and dropped.

AI doesn't need to invent that idea. It only needs to make it cheap enough that the filter reconsiders.

Nobody suddenly got more creative. They stopped rejecting ideas they already had.

That distinction matters, because what looks like an explosion of new ideas is partly an explosion of suppressed ones coming back.

Once one of those ideas becomes real, the thing accelerates.

Finishing a project doesn't only produce a product. It changes how the next opportunity looks. You learn patterns, you keep components, you find three related problems you didn't know existed, and you believe the next one is achievable.

So the second project costs less than the first. Lower cost pushes more ideas through the filter, which creates more experiments, more reusable parts and more knowledge, which lowers cost again.

More capability leads to more attempts, and more attempts create more capability. That isn't higher productivity. It's accelerated exploration.

Entrepreneurs hit this early because they live at the execution limit.

They don't have more ideas than everyone else. Almost everybody has ideas about what could be built, automated or organized better. Entrepreneurs are simply the ones who keep testing the boundary between an idea and a working thing.

Historically even the productive ones ran into a wall. Fifty promising ideas, enough time and money for two. The rest stayed in notebooks or evaporated.

AI moves that wall. One person can now test more concepts, build more prototypes, probe more markets and kill weak ideas earlier. The portfolio of realistic options expands, not just the speed.

Which creates a new problem. You stop running out of implementation capacity and start running out of attention.

Yesterday the bottleneck was implementation. As implementation gets cheap, selection matters more. The hard question stops being “can I build this?” and becomes “should I?”

Then it moves again. Suppose AI produces ten finished prototypes overnight. Your job isn't building anymore, it's deciding which of the ten deserves another day. Which solves a real problem. Which creates value. Which is worth maintaining, and which should be shot.

Eventually AI helps with that too — measuring engagement, running the experiments, comparing outcomes, proposing alternatives, recommending a direction. So the bottleneck shifts once more, from judging solutions to defining objectives worth having.

People increasingly specify goals rather than implementations. Instead of writing software they describe outcomes; instead of designing every workflow they define what success looks like.

The work moves up the stack. Scarcity survives the whole journey — it just migrates from execution to judgment, attention, priority and direction.

Most AI conversations are about productivity. Hours saved, software made cheaper, tasks per person, headcount avoided. Fair questions, and they miss the bigger effect.

The largest change may not be people doing existing work faster. It may be people attempting work they'd previously written off.

History rhymes here. When photography went digital people didn't take the same number of photos more cheaply — they took vastly more. When cloud computing got cheap, companies didn't merely cut infrastructure bills, they launched products that would never have justified buying servers. When publishing went digital, production reached people no traditional publisher would have touched.

Every one of those reduced the cost of execution, and every one produced an explosion of experiments, because projects that used to fail the internal filter suddenly passed it.

AI may push harder than any of them, because it cuts the cost of thinking, planning, designing, coding, writing, researching, analyzing and iterating simultaneously. A cheaper server bill changes one part of a project. This changes almost all of them.

Jevons explains why cheaper AI leads to more AI. The second-order effect is that cheaper execution changes which ideas survive the filter at all.

Lower cost doesn't only raise output. It raises the number of ideas that live long enough to be tested.

So a society doesn't simply produce more of the same things. It explores a larger share of the possibility space: more experiments, more startups, more niche products, more hypotheses, more creative work, more attention on problems nobody bothered with.

Most of those fail. That's the point of trying them. What matters is that they exist at all, where before they'd never have been attempted.

Testing ten times as many ideas isn't working ten times faster. It's raising the odds of finding the ones that were invisible.

It might be a mistake to file AI under automation. It's more useful as a technology that lowers the threshold between imagining something and building it.

For most of history ideas were abundant and execution was scarce. That balance is inverting. As execution gets cheap, the scarce things move: attention beats implementation, judgment beats production, and picking the right objective beats working out how to hit it.

The value chain moves upward. Less time turning ideas into reality, more time deciding which realities are worth having.

That doesn't make implementation irrelevant. It makes implementation stop being the thing that decides which ideas get tested.

The filter stays. Its thresholds move. Ideas that looked absurd become merely expensive, expensive ones become practical, practical ones become trivial — and ambition adjusts to each shift.

Jevons is still a good lens. Cheaper, better AI means we'll use much more of it.

But the deeper change is psychological. Your brain filters ideas by what it believes they cost, and when AI drops that cost, ideas that used to die on contact start surviving. Some become experiments. Some become products. Some become companies.

The biggest impact of AI may not be humanity doing its existing work faster. It may be humanity finally exploring a decent fraction of the ideas it already had.

Most will fail. A few will change industries. All of them exist because the filter gave them a chance it never used to.

That isn't only a productivity revolution. It's also a possibility revolution.