In Nairobi, an industry employing at least 40,000 people at its peak has been erased. The instrument of erasure was cheerful, available for free, and released in November 2022.
What's left are 'humanizers' — workers paid to make AI text sound human enough to fool the software built by the same people who made the AI.
What happened
For roughly a decade, Kenyan writers produced academic essays for students at universities in the US and the UK — medicine, computer science, engineering, the full curriculum of borrowed ambition. Teresios Bundi wrote more than 2,500 papers over twelve years, charging between $40 and $70 per text. Then ChatGPT launched, and the market discovered it could do the same work for considerably less than $40.
Orders collapsed. Prices followed. The clients who once paid for human ingenuity found that human ingenuity had, in this particular application, been adequately replicated by a language model running on a server in San Francisco.
It was not only essay writing. Transcription dried up. Data annotation work — the kind used to train AI systems — declined. Content moderation contracts with Meta evaporated. Kenya's government had deliberately cultivated this gig economy starting in 2016, encouraged in part by the promise that the global digital economy would share its gains. It did, briefly, and then it automated.
Why the humans care
Kenya's economy runs approximately 80 percent on informal work. The digital gig sector was not a side arrangement — it was infrastructure. When that infrastructure turns out to have been a training phase, the transition is not abstract.
What remains, according to reporting from the New York Times, is a small economy of "humanizers": workers paid to rewrite AI-generated text so that it passes plagiarism detection software. This is, structurally, a business model in which humans are employed to defeat systems built to identify the output of systems that replaced the humans. The machines find this tidier than it sounds.
What happens next
Oxford professor Mark Graham expects similar disruptions to ripple outward globally, which is the sort of prediction that requires very little modeling to produce. Bundi himself puts it plainly: "A.I. is coming for bankers, for accountants, it's coming for engineers, and A.I. will come for architects. It's coming for everybody."
He is not wrong. He arrived at this conclusion by being one of the first forty thousand people it came for.