Six workers show what the real AI economy looks like below the frontier-model headlines

Rest of World followed six workers using AI and found a more informative picture than adoption surveys provide. A Lagos makeup artist uses free ChatGPT as a hiring manager, sales coach, pricing calculator and content assistant; she estimates about 5% business growth, but briefly stopped when dependence became uncomfortable. A Bengaluru café owner pays $26 a month for agents that calculate recipe costs including labor, electricity and overhead, compare local prices and translate staff instructions. She trusts AI with arithmetic on her own data, not with emotional judgment or customer profiling.

At the disruptive end, a Zhengzhou studio says an AI-generated mini-drama can cost about $20,000 with 10 people in under a month, versus roughly $200,000, 50 people and two months for live action. The work has not vanished so much as changed into repeated sampling: because a usable generated second costs around 0.8–1 yuan, workers may generate 4.5–6 hours of footage to assemble a 90-minute series. Traditional crews have largely left that production line.

Elsewhere, a Mumbai dance-school founder used Suno to produce 60 songs in three months after a studio quoted about $629 and three months for just 10; a Taiwanese trainer with no coding background built a Claude app that freed time for three additional clients; and a Corteva scientist says AI-assisted analysis may eliminate a second year of some pesticide field trials, while cutting performance-review writing from roughly three weeks to one.

The cases are anecdotal, not a representative productivity study. Their value is mechanism: AI pays first where it compresses coordination, calculation and low-budget content production, while embodied judgment and confidential work remain harder to delegate.

The early labor effect is neither universal replacement nor mere autocomplete — it is a widening gap between workers who can bundle many functions into one role and workers whose previously scarce output can now be generated cheaply.