A professional cleaner arrives, scrubs your kitchen, vacuums your floor, and leaves without charging a cent.
The catch is a head-mounted camera recording every move. That’s the deal that startup Shift launched in New York City last week.
Demand reached thousands of bookings within hours of going live, Semafor reported Friday (May 29).
Shift doesn’t make money cleaning apartments. It makes money selling what the camera sees.
Shift is an offshoot of Germany-based Microagi, which already oversees data collection in several countries, anonymizing footage and licensing it to AI labs, the report said.
The value isn’t the square footage. It’s the clutter on the counter, the awkward stack of dishes, the stain in the corner that a robot must learn to find. Real homes carry that kind of unpredictability. Staged lab videos don’t.
Advertisement: Scroll to Continue
The Data Problem Robots Can’t Fake
Language models trained on the web had it easy. Decades of digitized text, images and video were already there. For robots, that supply doesn’t exist.
A robot learning to wipe a counter needs multidimensional sensor traces like vision, force, joint position and motor command to be captured in tight time synchronization during a real physical interaction, Tech Times reported May 16.
Simulations struggle to model physics accurately enough for grasping and moving objects in real-world conditions, MIT Technology Review reported April 1. So, they’re being trained on movement data.
In China, workers wear exoskeletons and virtual reality hardware to show robots how to open a microwave and clean a table, the report said. Meanwhile, gig workers in Argentina, India and Nigeria film themselves doing chores at home.
The industry’s data problem has spawned its own global labor market.
Shift’s approach differs. Rather than paying workers to record themselves, it subsidizes the labor directly, hiring vetted cleaners, equipping them with camera rigs, and absorbing the full cost of the service.
The company’s bet is that first-person footage of real household tasks carries enough value to cover a cleaning session, Cloud News reported. The dirty apartment is worth more as a data point than as a job billed conventionally.
A New Market for Physical Behavior
Shift isn’t alone. In March, DoorDash launched Tasks, a standalone app redirecting its 8 million delivery couriers in the United States toward generating training data for AI and robotics systems. Assignments include loading a dishwasher, folding clothes and filming unscripted conversations. DoorDash uses submitted footage to evaluate its in-house AI models and those built by partners across retail, insurance, hospitality and technology.
Uber launched a comparable pilot in late 2025, paying U.S. drivers to upload photos and recordings through its AI Solutions.
These platforms share a workforce already dispersed through people’s homes and everyday routines, with the infrastructure to direct it toward data capture at scale.
From Free Services to a Data Exchange
Shift plans to expand into plumbing, cooking and building repair, trades that have resisted automation because the physical complexity is harder to simulate. Each new service category is a new dataset.
Microagi paid more than 10,000 operators across 15 countries over $5 million in the first quarter of 2026, Entrepreneur reported Monday (June 1). The free NYC cleanings are a promotion meant to enlist more contributors.
Consumers once traded browsing behavior and social graphs for free email and social networks. A similar exchange is taking shape in the physical world: access to real-world human activity in return for services that would otherwise cost money.
For robotics companies, the next defensible position may not be model architecture or hardware design. It may be proprietary real-world training data.
For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.
