The Rise of Egocentric Data: How Gig Workers Are Training Humanoids for Pennies

The Growing Market for First-Person Robot Training Data

A new labor market is emerging at the intersection of gig work and artificial intelligence: companies building humanoid robots are purchasing vast quantities of egocentric video—first-person footage of people performing household chores. According to Wired AI, investors estimate leading robotics firms will acquire hundreds of millions of hours of this specialized video content over the next few years to train machines on fine motor tasks like dishwashing, laundry folding, and pouring drinks without spilling.

The demand for egocentric data has created a cottage industry of platforms recruiting everyday workers to record themselves completing household tasks. Although the internet contains abundant video content, the specificity required—thousands of close-up angles showing the same motion from a first-person perspective—makes scraped footage insufficient for real-world robot training. This hyperspecific requirement has opened a niche but growing market.

From DoorDash to Specialized Platforms

DoorDash, the food delivery company, launched a stand-alone application called Tasks earlier in 2026 to formalize this data collection pipeline. According to Wired AI, the move signals broader industry recognition that egocentric data collection will become a meaningful labor category in the United States.

However, other platforms are moving faster. Kled, a data marketplace founded by 22-year-old Avi Patel, explicitly aims to scale the work. Patel told Wired AI: “I want every person on the planet to be recording themselves doing the dishes. That’s going to make a robot so that you never have to do the dishes ever again.” Additional platforms including Luel and Waffle Video are also recruiting workers, creating competition in a field that barely existed two years ago.

Economic Reality: Minimal Earnings in High-Cost Markets

The financial reality for gig workers in the United States is bleak. According to Wired AI’s firsthand account, a full week of dedicated household-task recording yielded negligible compensation—far below the $2,500 monthly rent in San Francisco. The reporter signed up for three platforms (Kled, Luel, and Waffle Video) after testing DoorDash’s Tasks app and found the earnings insufficient to meaningfully offset housing costs.

The wage dynamics differ starkly by geography. In India, where self-employed workers typically earn around $125 per month on average, egocentric data gigs offer comparable rates, making them more attractive in lower-wage labor markets. This geographic arbitrage is already driving growth in countries outside the United States, where the labor supply is more abundant and cost expectations are lower.

Why This Matters

The emergence of egocentric data collection as a scaled labor market reveals a structural tension in AI development: the most advanced humanoid robots require hyperspecific, labor-intensive training data that cannot be sourced from public internet archives. This creates demand for workers willing to perform repetitive, recorded household tasks—work that is economically rational in low-wage countries but economically irrational in high-cost US cities.

For robotics companies, the math is simple: hundreds of millions of hours of training video, priced near-zero in arbitraged labor markets, is a bargain compared to accelerating robot capability timelines by years. For US gig workers, the proposition is far less attractive, which suggests the real growth in this sector will remain concentrated in India and other lower-wage economies. The irony is sharp: humans are training the robots that will eventually eliminate the very tasks—and task-related gig income—that made the training economically viable in the first place.