Platform monopolists extract rent from exhausted workers by selling the illusion of effortless escape.
It is true that a mechanical engineer in Austin can make fifty thousand to a hundred and fifteen thousand dollars a year selling an oversized lint roller on Amazon while working two hours a month. It is true that a paper-mill worker outside Montreal can use an AI chatbot to find a hyperspecific gap in the Etsy search index — say, a meal planner for women with ADHD who hike — and sell a PDF he generated in minutes. It is true that a 31-year-old American in Spain can upload two hours of his speech to ElevenLabs and earn about $3,000 a month licensing his AI-cloned voice for audiobooks, guided meditations, and fantasy narration. In the narrow sense in which a working system is a working system, these are successful arbitrage trades. They exploit the platform’s initial distribution of attention to extract a toll-free surplus. And that is exactly the vulnerability.
The platform-political-economy literature has a name for what happens when every worker realizes the same arbitrage is available. Cory Doctorow calls it enshittification. The platform is good to its users long enough to lock them in, then good to its business customers, then claws the value back from both. The lint-roller maker and the Etsy PDF-peddler are operating in the third stage. They are not escaping the wage-labour trap; they are temporarily occupying a crack in the floorboards before the platform’s search-ranking apparatus notices the overcrowding and prices the rent back up. The algorithm is not a magic money-generating wand; it is a continually-tuned set of weights serving a continually-revised objective function for the platform’s shareholders — a mechanism that adjusts search rankings and visibility per-user, per-second, at machine speed to extract maximum transaction value.
The dream is rational — and that is what makes the racket dangerous. In March, the share of U.S. workers satisfied with their pay and opportunities for promotion hit its lowest level since the New York Fed started asking the question in 2014. A survey commissioned last year by the investing platform dub found that 60 percent of Gen Z adults believe a conventional full-time job will not let them reach their financial goals. Around one in four Americans now report a side hustle. The labour market is, by every available indicator, performing a kind of slow extraction on the people who show up for it, and the people who show up for it have begun to look for exits. Older Americans turning to gig work in retirement to cover bills is the same arithmetic on the other end of the age curve. Passive income is the vocabulary the exits arrive in. The trouble is what gets built around the vocabulary.
Start with the courses. Ana Lohrmann, a 43-year-old former Spanish teacher in Catonsville, Maryland, who is in treatment for cancer and cannot work a regular schedule, has sunk thousands of dollars into courses run by entrepreneurs who promised she could make money in her sleep. A $1,000 course on publishing email newsletters turned out to be too general, so she paid $2,500 for a more specific one. “I don’t know of anybody who made as much as they were hoping to make or as much as the teachers of these courses say you’ll make,” she said. The median loss reported to the Better Business Bureau last year among people who said they had paid for business courses that did not pay off was $1,326.
What is new, in 2026, is that an AI chatbot will coach you into the schemes. Lohrmann, in Maryland, asked one for an idea. It proposed a Spanish-language placement test, sold on a classroom-resources website, and told her she would likely make $7,000 in her first year. She spent a few weeks on the project. About a year later, it has brought in about $250. The chatbot did not lie, exactly; it produced, with the cheerful confidence of a next-token predictor trained on the same Reddit threads and YouTube testimonials Lohrmann had been reading, the median outcome dressed up as her expected one. The asymmetry is structural: a system trained on the promises of passive-income schemes will reproduce those promises, with no penalty for being wrong and no ability to know that it is wrong. The model does not know what Etsy shoppers are searching for. It knows what people who wanted to be Etsy shoppers said they were searching for, which is a different thing.
Then there are the platforms themselves, which are doing more than they say. ElevenLabs began licensing people’s voices in early 2024. The company says it has paid out $22 million to more than 10,000 uploaders since the program launched. That works out to an average of about $2,200 per uploader over the lifetime of the program so far — a sum that, annualized, sits well below the federal poverty line for a single adult, in case anyone is keeping score. ElevenLabs is not running a passive-income business for its uploaders. It is running a data-acquisition business that pays a small fraction of the rent it will eventually extract from the resulting models, in the same way that the early Uber drivers were not running a gig-economy business for themselves but were providing the demand data Uber needed to price the market for its own benefit. The uploaders are an input to a model, paid at the rate of inputs.
Ronnie Lim, a 19-year-old in Lawrenceville, Georgia, has found a more honest version of the same game, though he has not framed it that way. He makes thousands of dollars a month by running eBay stores that list products at a markup and ship them from Amazon when the order comes in. He says he does not understand why his customers buy from him on eBay at the prices he charges. The reason is that eBay’s search algorithm rewards stores that ship fast, and Amazon’s two-day delivery is fast, and the markup is a rent he is collecting on the gap between two platforms’ logistics. He is not making money from passive income. He is making money from a structural arbitrage between two platforms whose terms of service both forbid what he is doing and neither of which appears to be in a hurry to stop him. Amazon says buying products on its platform to resell them is against the terms of its Prime service. eBay forbids dropship-fulfilment. Lim shrugs. “We don’t know how long this is gonna last but we’re just gonna make as much money as we can.” The Justice Department, in a recent case, charged a North Carolina man who used a bot army to stream AI-generated songs billions of times, racking up more than $8 million in royalties. He pleaded guilty in March to conspiracy to commit wire fraud — which is what happens when nobody at the platform bothers to ask whether the music is being listened to. The pattern, in every case, is the same. The platforms are paid, in some mix of fees and plausible deniability, to look the other way.
Greg Keogh, the Austin mechanical engineer who nets $50,000 to $115,000 a year from a wide lint roller he designed and listed on Amazon, is the closest the catalogue has to a real success story, and even his success story includes the line: “The more pain you have in the beginning, the more passive it might be.” Keogh spent a Saturday in his garage swapping 1,600 faulty handles out of his first production run. He decided not to expand into Target and Costco. He now works more hours a week than he did when he had a job. The lint roller is, in the technical sense, rent he collects on a product he designed once and will not have to design again, and there is nothing wrong with that, but it is not passive income. It is a small business that happens to require less of his time than a job. Most of the schemes in the catalogue are not even that. They are what an older generation of Polish-Canadian extraction victims would have called, in a vocabulary unavailable in polite magazine writing, making a living from making other people think you are making a living.
J.K. Galbraith called the related phenomenon the bezzle — the magic interval when the fraudster has his gain and the victim does not yet understand he has lost it. In this casino economy, the victim and the fraudster are the same person split across two moments: the seeker believes they are escaping wage-labour through self-deception, even as the platform extracts the structural surplus they generate. The profile of a nineteen-year-old running eBay arbitrage stores, or a software-tutorial maker licensing his cloned voice for audiobooks, reads less like a new economic paradigm than like the gravity-defying moment before the collapse. When the young are left doomspending as economic anxiety grows, the most lucrative product in the casino economy is not the side hustle itself; it is the course teaching the side hustle.
The remedy is structural, and the playbook is older than the mechanism. A bar mill in Selkirk, Manitoba, bought by the Brazilian conglomerate Gerdau in 1995, sped up its line to extract the surplus the existing workforce had built, locking the workers in by raising the cost of leaving. The mechanism today is cloud infrastructure, API locks, and algorithmic wage discrimination. The digital platform is not a new economy; it is the latest extractive-staple political economy operating from the metropole, treating the creative and logistical labour of its users as a free input to be harvested before the arbitrage window closes.
The Federal Trade Commission has been doing what it can. In 2022, the agency secured $2.8 million in refunds for 890 consumers of an “autopilot” dropshipping scheme, and it has separately shut down a semi-truck leasing scam targeting gig workers. The Justice Department has prosecuted the bot-streaming operator. But enforcement is downstream of an economy in which the platforms are paid to look the other way, the AI tools are paid to encourage the schemes, and the course sellers are paid to sell the dream regardless of whether the dream works. The more durable answer is to wield the FTC’s Section 5 unfairness authority to ban the algorithmic twiddling that hides the pea in the shell game, to deploy DMA-style interoperability mandates and the American Innovation and Choice Online Act to break the monopsony chokepoints, and to write a federal privacy law with a private right of action so that the people whose data and labour feed the machine have standing to sue. Right-to-repair legislation worked because farmers who owned half-million-dollar tractors could not, in the end, be forbidden to fix what they owned. Interoperability mandates worked because the platforms could not, in the end, lock the data in. The same architecture is going to have to be built for the income-extraction layer that the AI buildout is now assembling on top of the old labour market, because the workers who show up for it are showing up for a market that has been re-engineered to extract from them, and the exits that get sold to them are, mostly, just the next room in the same building. Under that regime, Lohrmann is not left to navigate the wreckage of her own bad bets; the regime simply outlaws the sale of the delusion to her in the first place, treating the platform’s extraction of her desperation as the primary harm rather than an unfortunate cost of doing business.
Keogh, the lint-roller magnate, knows what the tradesman knows — that the only thing that makes a machine run without supervision is the quality of the initial machining. But the platform did not machine the machine. The platform owns the floor you are standing on, and it will raise the toll the moment you prove the floor is worth standing on. The dream is rational. The schemes mostly aren’t. The platforms, the AI tools, and the course sellers know that, which is why they are all still in business. A Polish saying, which translates badly, has it that the work doesn’t care how you feel about it. The work is to be done.
— Stewart Letterkenski