The Jevons Paradox for AI: Why "It's Too Late" Is Exactly Backwards

You've heard it: "AI is everywhere now, the market's saturated, you missed it." It sounds reasonable. If a tool gets cheap and good enough that anyone can use it, surely the opportunity dries up. That intuition is wrong, and there's a named economic law that says so. Once you see it, "it's too late" stops sounding like caution and starts sounding like an excuse.

• The fear: cheaper + better AI = less work for people who build with it.
• The reality: when a useful thing gets cheaper, we don't use the same amount more cheaply. We use far more of it.
• The name for that reversal is the Jevons paradox, and it's been holding up since 1865.

In 1865 the economist William Stanley Jevons noticed something odd about coal. Better steam engines used less coal per unit of work, so everyone assumed Britain would burn through less coal. The opposite happened. Because steam power got cheaper, factories, trains and ships used vastly more of it, and total coal consumption shot up. The efficiency didn't shrink demand. It unlocked uses that weren't worth it before. That's the whole paradox: when something gets cheaper to use, falling price can grow total demand instead of cutting it.

• Efficiency lowers the cost per use.
• Lower cost makes things worth doing that weren't worth doing before.
• Those new uses can outweigh the savings, so total consumption rises.

The cleanest modern example has nothing to do with chatbots. ATMs were supposed to kill the bank teller. They didn't. Economist James Bessen documented it: as ATMs spread, each branch needed fewer tellers, so banks opened far more branches, and the total number of tellers actually rose. Cheaper-to-run branches meant more branches, which meant more teller jobs, not fewer. Hold that shape in your head, because AI is running the same play right now.

On June 9, 2026, Andrej Karpathy — now at Anthropic — reacted to a major model release by naming the same effect for software directly. As he put it, working software increasingly "comes out on a tap," and so "the Jevon's paradox kicks in" and his own "demand for software [is] growing substantially." That's the tell. When the person who can build anything finds himself wanting MORE software, not less, the market for building isn't closing. It's opening. Cheap creation doesn't satisfy demand for custom software. It exposes how much was never built because it cost too much.

• When code is cheap, the bottleneck moves from 'can we build it' to 'what's worth building'.
• Every dashboard, internal tool, and bespoke single-use app that wasn't worth a developer's week is suddenly worth an afternoon.
• That backlog is enormous — which is exactly the elastic-demand condition the Jevons paradox needs.

Read carefully, because the paradox doesn't promise everyone wins. It says total demand grows — it doesn't say it gets handed out evenly. When coal got cheap, the winners weren't people who admired steam engines. They were people who built mills, railways and ships on top of cheap power. Same now. Cheap AI rewards the people who turn it into something a specific buyer wants, not the people who just have access to it.

1. Stop treating access as the advantage. Everyone has access. The advantage is knowing one audience's problem cold.
2. Pick a narrow 'now-worth-doing' job — something nobody bothered automating because a developer was too expensive for it.
3. Build the thin layer between the cheap model and the buyer: the workflow, the packaging, the support. That layer is where the money sits.
4. Ship the small version this week. The paradox rewards motion; 'I'll wait until it settles' is how you opt out of a growing market.

When something genuinely useful gets cheaper, humanity doesn't buy less of it — it finds a hundred new reasons to buy more. AI is the cheapest it's ever been and the best it's ever been, at the same time. 'Saturated' and 'too late' are descriptions of a shrinking market. This one is doing the opposite.

• Cheaper + better → more total demand, not less (when demand is elastic).
• The builder/reseller market is widening, not closing.
• The risk isn't being late. It's mistaking 'everyone has access' for 'everyone has an offer.'