In New York City, where almost nothing is free, a startup called Shift Robotics is offering to clean your apartment for $0.
There is, of course, a catch: The cleaners wear head-mounted cameras that record first-person footage as they do dishes, mop floors, and fold laundry. That footage is turned into training data for AI labs and robotics companies.
Few people had heard of Shift before last week. Then its launch video went viral, drawing more than 8 million views. The initial 250 cleaning sessions sold out almost immediately. “We had thousands and thousands of people trying to book,” Harry Kilberg, Shift’s US General Manager, told Business Insider.
Shift is part of a growing scramble for real-world data as AI companies move from chatbots to machines that can work in homes, warehouses, and factories. Sam Altman has declared robotics OpenAI’s next frontier, while Nvidia, Meta, and Tesla are all ramping up their own robotics efforts.
Kilberg, who stars in the launch video, said he expected the frenzied response. “We knew this idea is world-changing, so we thought it would go viral,” he said.
Shift is the consumer-facing brand of microagi, a research lab founded in a Munich hacker house last year. The lab, headquartered in Germany, says it is working on “end-to-end physical AGI,” or artificial general intelligence for machines that can act in the real world. microagi was founded by Bercan Kilic and Yoan Iliev, both former Formula One aerodynamic engineers, and Anton Poletaev, a former researcher at The Alan Turing Institute.
Shift operates in 15 countries and has 14,000 operators collecting real-world data, according to Kilberg. He described it as a marketplace “aimed at accelerating the transition from an economy where people work out of necessity, to one where everyday goods and services can become increasingly abundant and accessible.”
Kilberg said robots that can reliably handle household chores are still far off. Until then, he said, people can be paid to gather the examples, aka the data, that those systems need to learn.
That data has become one of the biggest problems for startups and Big Tech companies trying to move AI from chatbots into machines that can act in the physical world. Large language models are trained on vast amounts of text and images from the internet. Since a comparable corpus for robots does not exist, the industry is trying to create one from scratch by paying gig workers to record the work robots may eventually do.
The free cleaning offer raises an obvious question: Does the math work?
“The unit economics are a lot better than you think,” Kilberg said. microagi’s internal technology processes the data in a way that makes it significantly higher quality, allowing it to command a premium when sold to AI labs and robotics companies, Kilberg said. Faces and screens are automatically blurred in the videos to anonymize the data, and no audio is captured. microagi also uses the data for its own internal research.
Kilberg said the idea for the launch came from early users who were already recording themselves doing tasks around their homes and wanted to do more.
“They started putting up flyers in their apartment buildings offering cleaning to neighbors because we would foot the bill,” Kilberg said.
Others began stocking shelves at local bodegas or volunteering at soup kitchens while recording the work, he said. Kilberg declined to say how much people are paid to record themselves doing chores.
New York is just the beginning. Kilberg said Shift plans to expand across the US and add free or subsidized services beyond cleaning, including cooking and plumbing.
Shift is part of a growing market for data collected from the physical world. Scale AI, Turing, and micro1, which helped supply data for the chatbot boom, have moved into real-world data collection. The goal is to help close what UC Berkeley roboticist Ken Goldberg has called the “100,000-year data gap”: the idea that robots are far behind chatbots because they lack enough real-world training examples.
The more services Shift offers, the more useful its footage becomes. AI labs and robotics companies need training data from a wide range of tasks and settings so robots can learn from the messiness of the real world.
Shift is also focused on geographic diversity. Kilberg said the company operates in countries where few others are collecting this kind of data, including Bulgaria, Georgia, and South Africa. He said Shift has especially caught on in Turkey.
