In Villa Mercedes and Córdoba, a scene that just a few years ago sounded like low-budget science fiction already exists: workers with iPhones 11 and 12 strapped to their heads walk through factories, shops, and homes while recording hand movements, routines, and daily tasks. They aren’t filming an experimental series; they are training artificial intelligence. The company behind this is Atlas Capture, a subsidiary of Mecka.ia, which in October 2025 established its first South American base in Argentina to expand its global network of physical data.
The method is both simple and sophisticated, a frequent combination in the digital economy. The company hires local recorders, provides the equipment, and assigns them egocentric captures: the camera records from the worker’s perspective, focusing on hands and actions, never on faces or personal data. Then, that material is reviewed to remove sensitive information, and each participant signs confidentiality agreements. Put elegantly: controlled privacy. Put bluntly: someone cleans hours of video so a machine can learn to open drawers without an existential crisis.
The scale is not small. Atlas Capture handles monthly projects of between 5,000 and 20,000 recording hours, has 50 employees and 17 local associates, and has already visited more than 160 establishments including factories, businesses, and homes. This volume feeds computer vision systems, task perception, and object manipulation destined for domestic robots and humanoid industrial machines. That is to say, it’s not about teaching a chatbot to write friendlier emails, but about teaching a machine to understand the physical world, where everything has weight, falls, and breaks things.
According to Lautaro Porras, the goal is not to replace workers, but to facilitate and improve their work. He also maintains that the project generates real income for local people and businesses. Both statements can be true at the same time, which makes the matter more interesting and less comfortable.
The Argentine case fits into a global trend known as “ghost work”: the invisible labor that sustains AI. Before, it was labeling images or moderating content; now it also involves recording how fruit is peeled, how a box is arranged, or how a tool is used. Mecka.ia already operates in Toronto, Bali, Vietnam, Kenya, and the Philippines, while in the United States, similar captures with GoPros and headbands proliferate in urban homes where there is always someone willing to monetize the future.
In parallel, the demand for AI trainers skyrocketed in 2025. Platforms like Deel reported explosive global growth, with Argentina and Colombia among the most dynamic markets. In the country, these jobs pay around 20 dollars per hour, with weekly payments and a promise of flexibility. Not bad for an economy that usually offers uncertainty for free.
The final irony, of course, remains intact: today these are real jobs, better paid than many traditional ones, but they produce exactly the data that will allow those tasks to be automated tomorrow. Technological capitalism rarely loses its sense of humor.
If you’re interested, I got it from here: