Kenya's essay industry died in two years. The jobs plan didn't notice

The most useful AI-jobs story of the year is not about engineers or call centers. It is about an industry nobody was supposed to defend, and a government plan that assumed work like it would keep growing.

What actually disappeared

A New York Times feature published Saturday counts the collapse: at its peak, more than 40,000 people in Nairobi made a living writing essays and coursework for students in the United States and Britain. Two years after ChatGPT, the trade is largely gone. Students made the calculation the Times describes plainly — why pay a freelancer in Kenya when the chatbot does it?

The reporting follows one worker through it. Teresios Bundi moved to Nairobi from a farming village in 2011, earned $7 for his first three-hour assignment — a two-page essay on a fruit he had never heard of — and went on to write more than 2,500 essays over twelve years, sometimes three a day. He finished a public health degree in 2015 and never used it, because the essay business paid at least five times as much. He rented a house near a university, filled it with desks, fast internet and mattresses, and charged $40 to $70 a paper. He has since shut the operation down and works for a German development agency helping young Kenyans find digital work. Richard Esilaba, who employed 100 writers at his peak, has closed. Writers who cleared $900 to $1,200 a month now report $500 to $800.

This is not a story about a skill that turned out to be worthless. It is a story about a skill that was real and a market that only ever valued it because it was cheap.

The plan was built on this kind of work

Kenya's essay economy was the informal edge of something official. Cheap high-speed internet reached the country around 2009, and the state came to see online outsourcing as a ladder for graduates — transcription, data entry, design, basic coding. Nobody condoned essay writing, but it fit the shape. In 2016 the government began training graduates to work the freelance platforms. In 2022 it adopted a ten-year National Digital Masterplan that leaned further into outsourcing and gig work. OpenAI released ChatGPT the same year.

The numbers behind that bet are not small. More than 100,000 students graduate from Kenyan universities each year, roughly 80% of work in the economy is informal, youth unemployment runs above 25%, and about 40% of the country's 54 million people live in poverty. John Tanui, the principal secretary at the ministry responsible, told the Times he still stands by Kenya's potential as an outsourcing hub. He may be right about the hub. The question is what work the hub will contain.

The essay collapse is the loud version of something that started earlier and quieter. Frida Mwangi went into transcription in 2015; from 2017 the work began drying up, because transcription was among the first jobs software could do. Her employers told her to run audio through AI and correct it, which she found slower than typing it herself — the models mangled medical terms. The jobs went anyway. Content moderation pulled back too: Sama cut back its Kenyan operations, and Scale AI discontinued some. We watched the American end of the same ladder close last month when Amazon pulled the plug on Mechanical Turk, AI's original human engine — the marketplace built because machines couldn't do these tasks, killed because they now can.

The number that says it is not finished

The most load-bearing figure in the story is not the 40,000. It is the automation rate. Scale AI and the Center for AI Safety run the Remote Labor Index, which measures how much real, paid freelance work leading agents can complete end to end. Last October the best models managed 2.5% of it. By July, 16%.

That spread is the whole argument. At 2.5%, the index's authors were right to say the data did not support panic: across 240 projects spanning 23 domains, the humans who originally did the work earned a combined $143,991 and the top-performing agent earned $1,720, with a median project worth about $200 and about 11.5 hours of human labor. At 16%, and climbing along that curve, the same sentence stops being reassuring.

But the shape of the work matters more than the rate. Agents in the index fail at complex editing, tool use and multi-step briefs. They succeed at generating from a simple prompt. An essay mill was always the second category — text in, plausible academic text out — which is why it died at 2.5% while architecture and game development projects were still safe. What remains in Nairobi is a grimly perfect illustration of the gap: the surviving jobs are "humanising," editing AI-written papers so they pass the detectors universities now run. One writer told the Times business is as good as ever; she drafts with AI and rewrites by hand, and says the students who get caught are the ones with no writing experience.

Mark Graham, the Oxford professor who studies the global gig economy, put the wider point to the Times: AI "has come in and taken big chunks of these labor markets," and Kenya will not be the only place.

The skeptic's case

Start with the obvious objection, because it is correct: this was contract cheating. Nobody owes anyone a living wage for helping students buy a degree, and the universities now running detection software are responding to a harm that predates AI by two decades. The industry was already low-trust and low-status, and the buyers were never loyal to the writers.

The counter that survives that objection is about the next rung, not this one. The ILO's 2024 survey of Kenyan online freelancers found academic writing was the single largest category of work, with article writing and editorial services close behind — meaning the country's digital labor market was concentrated in exactly the text-to-text commodity AI handles best. When the bottom rung is also the widest one, removing it does not just cost jobs; it removes the step people were standing on.

There is also a demand-side reading worth stating. This was not automation walking into a factory. It was the buyer leaving the market entirely, because the student can now do the cheating personally and for free. A worker displaced that way does not get rehired into a new task by the same employer — there is no employer left.

And the honest limit: 16% is not 100%. The Remote Labor Index says most complex freelance projects still fail for agents. What has been destroyed so far is a specific, thin, single-deliverable kind of knowledge work, not knowledge work as a category. Bundi's own line to the Times is the maximalist version — "it's coming for bankers, for accountants, it's coming for engineers" — and his own narrower arithmetic is the more useful one: the peers who stayed in public health have risen through the ranks, and if he went back now he would start at entry level under someone who graduated last year.

What to watch

Three things will settle whether this is a cautionary tale or a preview. Whether the Remote Labor Index keeps climbing at anything like the October-to-July slope, and whether the gains spread beyond generate-from-prompt work into the editing and tool-use tasks that still hold. Whether Kenya's outsourcing push — call centers, data work, content moderation, AI operations — actually moves workers up the value chain, or rebuilds the same commodity layer with better branding. And whether "humanising" survives the next generation of detectors, because a job that exists only to defeat a classifier has a shelf life measured in model releases.

The pattern we keep documenting is the same one from the other side of the ladder: AI is deleting the first rung, not the senior job traced it in Korean payroll data, where the losses piled up in exactly the entry-level, text-processing roles. Kenya is what that looks like when there is no senior rung to fall back on and the state's jobs plan was built on the step that just broke. We flagged the collapse this morning in Model fatigue is real: AI labs can't stop shipping, next to a week in which four labs shipped models nobody had time to evaluate. Both halves belong in the same picture.

Was Kenya's essay economy a real casualty of AI or just a bad market finally closing? Tell us in the comments.

Sources: The New York Times — Kenyans Did College Students' Homework for Years. Then A.I. Arrived. · The Next Web — Kenya bet its graduates on gig work in 2022. ChatGPT launched that year · Remote Labor Index (Scale AI / Center for AI Safety) · Scale AI — The Remote Labor Index: Measuring the Automation of Work