On the weekend of March 7th and 8th, Lovable was giving away credits for International Women’s Day. I am a woman who builds two products alone, late at night, in the hours after a full corporate day, once my children are asleep. A free weekend of building felt like a gift handed directly to me.
So I took it. For hours that ran together I moved between CareerMirror and GradLight, two products with different users, different logic, different open problems. I made progress on both at a speed that felt, in the moment, like proof that the constraint had finally lifted. The thing that always slowed me down was gone. I was going to use every minute of it.
Then the weekend ended, and I went back to look at what I had made.
Much of it I had to undo. Not because the tools failed. Because I had. I had made decisions on one product with my head still half inside the other. I had solved a problem in CareerMirror in a way that only made sense if you had been reading the GradLight code ten minutes earlier, which no one had been, including me by Tuesday. I had built quickly, I had built confidently, and I had built a great deal of work that my own attention, spread across two things at once, had quietly degraded without telling me.
Here is the part an economist cannot unsee.
When the cost of something falls, you do not bank the savings and carry on as before. You consume more of it. Cheaper coal did not lead to less coal. It led to mills and engines and a century of more coal than anyone had imagined burning. Economists call this the Jevons paradox, and it describes a free weekend on an AI building tool better than I would like. The price of building fell to almost nothing, so I did not do my usual amount and stop. I built twice as much.
But the cost did not disappear. It moved.
Building was never my only constraint. It was simply the one I could see. When AI made building almost costless, the bottleneck did what bottlenecks always do. It relocated to the next scarcest thing. And the next scarcest thing was not time, and it was not credits. It was my attention, which is fixed, biological, and does not come with a falling price curve. Every time I switched between the two products, I left a residue behind. A little of CareerMirror stayed lodged in my head while I worked on GradLight, and a little of GradLight bled back the other way, and the switching cost compounded in a currency I was not tracking, because I had been so fixed on the currency that had just become free.
I was not building more. I was building more poorly, faster.
This is why I do not believe in the weekend where someone builds twenty agents and announces it like a victory. I believe they built twenty things. I do not believe they thought twenty times. The artifacts pile up faster than the judgment that is supposed to govern them. And judgment, the work of knowing whether a given thing is correct, whether it fits, whether it should exist at all, runs on the one input that AI has not made abundant and likely cannot. Output and thought used to travel together, because building was slow enough that thinking kept pace on its own.
It does not anymore. The thinking has to be chosen now, deliberately, against a current that pulls the other way.
I still have the work I salvaged from that weekend. I also have the work I deleted, which taught me more. The credits were free. The bill came due somewhere I was not looking.
Abundance of output is not abundance of thought. It may be the most expensive thing we have stopped measuring.
— Irena