The TEMU-fication of Software, Digital Goods & Services

A hypothesis on the not-so-distant future of software, books, music, and movies, in which most of what we consume gets cheaper, more abundant, and noticeably worse, while the human-made variant moves into a luxury segment of its own.Disclaimer: This is an opinion piece and most of it is speculation about a future that has not arrived (yet?), based on a few data points that have. As usual, summary at the end.A few years ago I would have laughed at anyone telling me that there is a serious market for ten-dollar drills, two-dollar dresses, and one-dollar pairs of shoes shipped from a warehouse on the other side of the planet. Today, however, that market exists and it has a name, and it is even publicly traded (sort of, through holdings). TEMU, Shein and a few others have built frankly mind-boggling businesses around the idea that if you make production cheap enough, fast enough, and just barely good enough to look right on a phone screen, an enormous part of the population will buy it, even when the product breaks within a week, when the materials it is made of contain worrying levels of toxic substances, and when the carbon footprint of one delivery exceeds that of an equivalent local purchase by orders of magnitude.The key to this sort of business model is not innovation, but instead the externalization and compression of cost. Somewhere upstream, people work seventy-five hours a week, in conditions most readers of this website would refuse to even visit, so that the rest of us can have a cheap plastic spatula at our doorstep within five business days. While the visible price collapses, the invisible costs get distributed onto landfills, lungs, and ultimately people that we will never meet.What follows is a hypothesis I cannot prove but have been turning over in my head for a while, as we are watching the same thing happen to software, books, music, (film-)scripts, and most of the digital goods and services we consume.

And just like with physical goods, we will probably end up with a two-tier market, in which we have a large and massively profitable lower tier of generated slop, and a smaller, more expensive upper tier of work that is still recognizably human.I’d like to call this the TEMU-fication of software, digital goods and services, and describe what it might look like.Cheap laborFor decades, the global fashion industry has relied on a workforce that has almost no leverage and no voice, and for which the economics work because someone, somewhere far away, will sew a t-shirt for less than the price of a coffee. Without that skewed arrangement, the entire fast fashion business model collapses. The garment in your hand is only cheap to you because it has been expensive to someone else, in ways that the price tag does not show.Modern Large Language Models occupy a similar position in the economy, with one important difference, which is that there is no human being in the sweatshop, only a stack of GPUs trained on a corpus of work that other human beings produced over the course of decades. The labor that has been compressed is historical and the model is a kind of compressed copy of the work of millions of programmers, writers, illustrators, and musicians, served back at near-zero marginal cost.

The output passes the basic test of it runs and looks plausible, ships into production, and accumulates the kind of architectural and security debt that surfaces only when something goes very wrong.Note: There are credible voices in the industry, particularly from the AI tooling vendors themselves, who argue that AI-assisted development raises a floor more than it lowers a ceiling. In this view, the median piece of software has always been mediocre, written under deadline pressure by tired humans, copied from Stack Overflow without much thought, and held together by duct tape. If an LLM produces output of roughly comparable quality in a fraction of the time, the argument goes, nothing got worse. We are simply removing a bottleneck.I find this argument partially persuasive, and partially convenient for the people making it. It is true that a lot of software was already not great, but it is equally true that there is a difference between bad code written by a human who at least understood what they were doing, and bad code written by a system that does not understand anything. The first kind can be questioned and corrected, but the second kind tends to compound, because the person shipping it cannot answer why it does what it does. At least for now.Vibe-written booksSoftware is not the only place where this is playing out. The book industry is arguably further along, with estimates suggesting that somewhere between ten thousand and forty thousand AI-generated books are uploaded to Amazon’s Kindle Direct Publishing platform every month, many without any disclosure that a model was involved. In June 2023, the Kindle Top 100 bestseller list was found to contain only 19 books written by humans. Amazon has since introduced limits and disclosure requirements, but enforcement is patchy and authors continue to push back against what looks like a slow flood.Categories that have been hit particularly hard include travel guides (generated guides to cities the author has never visited, with restaurant recommendations that don’t exist), nutrition and health (generated diet advice with citations to studies that don’t exist), and public-domain rewrites (generated adaptations of older books, relying on the recognizability of titles that the actual authors never agreed to). Travel guides in particular have produced a small genre of stories where readers arrive at addresses that turn out to be empty lots, or follow walking directions through neighborhoods that no human would ever recommend.Note: The defense, again, is that the bottom of the book market was always full of filler, that print-on-demand has been around for a long time, and ghost-written business books and assembly-line genre fiction predate generative AI by decades. However, the new thing is the scale at which low-effort content can now be produced, and the speed at which it can drown out the rest of the catalogue.