She earns 250 rupees an hour teaching the robot built to replace her.

About 2.6 dollars. What an Indian worker is paid to film herself doing housework, so a humanoid robot can learn to do it. The pay band across the egocentric-data gig runs 250 to 400 rupees an hour; Human Archive's base rate is a single dollar.

Two hundred and fifty rupees an hour is the headline, but the transaction underneath it is the story. In a kitchen in Chennai, a 25-year-old films herself slicing mangoes with a smartphone strapped to her head, and that footage - first-person, hands-in-frame, the kind developers call egocentric data - is invaluable to companies teaching robots to move like humans. She is paid roughly 2.6 dollars for the hour. The robot she is training is being built to do the task she is performing. That is not a side effect of the deal. It is the deal.

India has quietly become the global middleman for the creation, processing and annotation of AI data, and the egocentric layer is the newest floor of that building. Objectways runs furnished fake apartments in Tamil Nadu where trainers record themselves folding the same towel ninety times a day; its subcontractor Qanat fits roughly 2,000 contributors with motion-sensor bands; Human Archive has put more than a thousand camera-caps into Indian homes, hotels and restaurants on an 8.2 million dollar round backed by angels from OpenAI, Nvidia and Google. The pay band sits between 250 and 400 rupees an hour. The work is real, it is growing, and a digital-labour researcher in Bengaluru expects it to grow further.

The number that frames all of this is not a wage, it is a warning. NITI Aayog's own roadmap says the AI conversation fixates on white-collar loss while ignoring India's 490 million informal workers - nearly half of GDP, ninety percent of the workforce, productivity barely 5 dollars an hour. Its blunt forecast is that informal income stagnates at 6,000 dollars by 2047 against a 14,500 dollar target, unless something deliberate intervenes. The egocentric gig is, for now, that intervention's awkward cousin: it puts cash in the hand today by selling the muscle memory that erases the task tomorrow.

This is the corridor read from the bottom. Last week the story was sovereign compute moving home and nobody hired to run it. This week it is the opposite end of the same stack - the people whose hands are being recorded so the stack has something to learn from. The capital, the chips and the buyers sit offshore. The labour, and the consent risk that comes with filming real homes, sits in India, Nigeria and Argentina. The wage is the signal. What it buys is the question the rest of the issue answers.

From labelling on a screen. -> To performing in your kitchen.

The old data-work bargain was the keyboard. You labelled images, you rated chatbot answers, you moderated content - work that lived on a screen and paid by the task. The new bargain is the body. Humanoid robots cannot learn to grasp, fold and pour from internet video alone, because simulations cannot model physics accurately enough, so the industry has turned to first-person footage of real people doing real chores. That shifts the work from the laptop to the living room, and it shifts the worker from annotator to performer.

Watch the money find the floor. Venture investors put more than 6 billion dollars into humanoid robotics in 2025; Goldman Sachs sees a 38 billion dollar humanoid market by 2035 and as many as 100,000 shipments this year. To feed that, data marketplaces - Scale AI, Micro1, Encord, DoorDash's Tasks app - are recruiting recorders across more than fifty countries. In the United States the pay is near-zero and the living costs are not. In India the same hour pays 250 rupees, and that gap is precisely why the cameras point here. The arbitrage that built the back-office decade has simply changed what it measures: not keystrokes, but gestures.

The same pattern is visible beyond India. Nigeria and Argentina appear in the same recruiting maps; South Africa's annotation rate sits at its own fraction of the Western wage. The Gulf enters from the other direction entirely - the UAE is now using AI and robotics to screen work-permit applicants, deciding who may come and work. So the region supplies the labour that trains the machine at one end, and deploys the machine to sort labour at the other. The corridor is not only compute and capital. It is people, recorded, rated and ranked.

The worker filming herself is not naive about the trade. She says she may get a robot herself one day; the studio graduate says the job is tolerable but she always feels the camera. That clear-eyed quality is the point. This is not a story about victims, it is a story about a price - what an hour of human demonstration is worth when the buyer is teaching a machine to need fewer of them. The seat that lasts is not the one in front of the camera. It is the one that owns the data, sets the consent terms and prices the gesture.

If the robots are coming, the durable seat is not in front of the camera. It is the human who owns the data, guarantees the consent and prices the gesture.

If the robots are coming, the durable seat is not in front of the camera, it is around it. The person who films a chore for 250 rupees is selling a perishable asset - once the model learns the gesture, the demonstration has no second buyer. The people who keep earning are the ones who own the pipeline: who source and manage the trainers, who guarantee the consent and the PII handling, who turn raw first-person footage into a clean, lawful, sellable dataset. Move up the stack from performer to owner before the performing work is automated by what it produced.

The second move is to make the consent the product. Egocentric capture films real homes and real bystanders, which means it carries real privacy exposure under GDPR, HIPAA-style health rules and the EU AI Act's high-risk-employment provisions landing in August 2026. A vendor who can prove clean, consented, auditable data collection sells something the cheapest recorder cannot. In a market racing on volume, the defensible position is verifiable provenance - the dataset a regulated buyer can actually put into production without inheriting a lawsuit.

The third move belongs to policy and to anyone building on top of it. NITI Aayog has named the 490 million and proposed Digital ShramSetu; the question is whether the egocentric gig becomes a bridge that lifts informal income or a treadmill that banks a few hundred rupees while training away the task. The difference is design - whether workers get a share of the dataset's downstream value, portable skills and consent rights, or a flat hourly rate and a smartphone on the forehead. The board that buys this data, and the founder that brokers it, both get to decide which one it is.

6 rising role categories, each with a sourced hiring signal.

The recording boom and the displacement risk are the same fact seen from two distances. Up close, egocentric data collection is new income for people the formal AI economy never reached - housewives, garland-makers, factory hands in Karur and Chennai earning by the hour. At arm's length, the footage trains the humanoids that the same Morgan Stanley and Goldman forecasts expect to number in the hundreds of thousands within years. The roles rising now sit in the gap between those two distances: the people who run, clean, govern and sell the data, not the people who perform it.

The named demand is concrete and it is local. Objectways and Qanat need supervisors and quality leads for studios and motion-capture floors in Tamil Nadu and Andhra Pradesh. Human Archive needs operations people to manage a thousand-plus headsets across Indian service businesses. The marketplaces routing this work - Scale AI, Micro1, Encord - need vendor managers who can source compliant data at volume. And the Gulf, hiring AI engineers at 31 percent growth and data scientists at 43, needs the governance layer that decides how screening and training data may lawfully be used.

The honest watchout is the one the workers themselves voice. This is perishable demand wearing the clothes of a growth industry. The performing gig pays today and trains its own replacement; the durable seats are the ones that own the pipeline and the consent. Chase the operator, quality, vendor-management and data-governance roles around the recording - not the recording itself. The hour in front of the camera is real money now and a closing window later. Build for the seat that is still there when the model has finished learning.