How ChatGPT Erased Nairobi's 40,000-Strong Essay-Writing Economy in Two Years

Here's What Actually Collapsed in Nairobi

• and Why It Matters More Than the Headline Suggests The story reads like a human-interest piece about one industry disappearing. The actual mechanism worth understanding is narrower and more precise: generative AI didn't destroy "jobs" in general - it destroyed a very specific kind of labour arbitrage and it destroyed it first, before anything else, because that arbitrage was built entirely on a task AI happens to do exceptionally well.

The Mechanism: What Teresios Bundi Was Actually Selling

Bundi wasn't selling expertise in Garcinia cambogia or engineering or medicine - he was selling the ability to produce competent, deadline-bound written text at a price lower than what it cost overseas students to write it themselves or hire help locally.

That's it. The value wasn't subject knowledge; it was cheap, reliable text generation. ChatGPT does exactly that task - generate competent text on demand - at near-zero marginal cost and without needing sleep, rent or a salary.

Once a free or nearly-free substitute existed for the single thing being sold, the arbitrage that supported 40,000 livelihoods in Nairobi had nothing left to stand on. It didn't shrink gradually because of quality concerns; it collapsed because the economic logic underneath it simply stopped existing.

Why This Segment, Specifically and Why First This is the

detail that makes the story more than an anecdote: cognitive gig work - essay writing, transcription, content moderation flagging - was exactly the category developing economies had used to leapfrog past the lack of local formal employment.

It required only internet access and language skill, not capital or infrastructure. That same low barrier to entry is precisely what made it the easiest category for AI to substitute, because a low-capital human task is, almost by definition, a task with few complicating variables an algorithm struggles to replicate.

The segment that let Kenya, the Philippines and India plug into the global digital economy without infrastructure investment is the same segment with the least protection against an AI substitute.

Where This Leaves Developing Economies Mark

Graham's warning - "it's going to be everywhere"

• is the right generalisation, but the specific lesson is about sequencing, not inevitability. The tasks most exposed first are routine, text-based, low-judgment cognitive tasks; the tasks that survive longest require contextual judgment, physical presence or domain trust that current AI can't yet replicate cheaply. For India, which has built substantial digital gig and outsourcing employment on exactly this kind of routine cognitive labour, the Kenyan collapse is not a distant cautionary tale - it's a preview of which specific job categories face the earliest pressure and a clear signal that reskilling strategies need to target judgment-heavy, trust-dependent or physically grounded work rather than more of the same low-complexity digital tasks that just proved most vulnerable.