~5 min read If you haven't heard someone invoke Jevons' Paradox yet, don't worry. You will soon, and someone will connect it with AI and coding, and I think they're just simply wrong. So first a definition: Jevons' Paradox is that as you as you make something cheaper to produce that has wide and general demand, like in coal in the original case, the demand for that thing increases. It's presented as a "paradox", and I frankly don't why it called called a paradox to start with. It follows supply and demand laws generally speaking. If you make something cheaper, demand in theory should go up, but when Jevons' first postulated this waaay back when. It was before the concepts of extensive vs intensive That said, Jevons' Paradox in more about how that as the price keeps going down for certain things that it seems we have unlimited appetite for. We have unlimited desire for more electricity, it seems. Also bandwidth and apparently lanes on a freeway. And the claim is now being extended out to AI. Ok, so here's what you came for: why do I say Jevons' Paradox related to AI coding is weird? It's weird because I don't think people are thinking very specifically about what specific economic units are being output. People invoking this usually think in terms of code. But that doesn't make sense. We in the industry know that lines of code as productivity is silly and terrible measure of productivity. You actually want some goldilocks of sufficiently short, but not too short in ways we don't have great attributes to measure what that means at the moment. And it's more weird to invoke Jevons' related to code, because AI is getting more expensive on a per unit basis, not less, in the near term. And yet people are generating more code—again—not less. Now lets say we're talking vaguely more software instead, first, I don't know how to count "softwares" but sure, I'll concede that there are more people who are able to make code can make features and products that didn't exist before. But I don't think counting softwares and saying those are getting cheaper is right either per se, because people are regularly excluding their own time to steer and AI to get things done. So maybe the measure is problems solved? And yes, we solve their problems than ever before and maybe we're reducing the cost of solving problems that people wouldn't solve otherwise. But another thing I want to observe is that we sort of knew that we have unlimited desire for things that solve our problems, and so letting people solve their own problems with new tools is obviously something that people will want. So let me state the weird bit plainly at the end: Jevons' paradox being applied to this coding itself is weird, because we don't have ways to measure the output or input. We can measure coal in and electricity out. We can measure components in for building out cables or towers and measure how much bandwidth it makes. But we don't really know how to measure this in software, and I don't know if we'll ever have it. So it's weird to apply this concept to a domain where we can't observe it clearly in my opinion. If you want to hear a longer and different take, you should watch John Green's take on Jevons' Paradox as it relates to AI and coding specifically. P.S. Dad Joke: What's a pirate's favorite programming language? Ye might say R, but a pirate's first love be C. This message is 100% organic human effort, so it will have typos. No, I didn't send it through an LLM to typo check it. It was made with just my fingers and my eyes. |