An AI company is sending actual humans with mops to scrub actual apartments in New York City for free. The stated goal: collect training data so robots can eventually do this job without the inconvenience of paying people.
This is not a thought experiment. This is happening. The company sends cleaners to your apartment, they clean it, and somewhere in a server farm a neural network watches the footage and learns what a squeegee is supposed to do. The homeowner gets a free apartment cleaning. The company gets unpaid labor disguised as customer acquisition.
Why does an AI company need humans to teach robots how to clean? Because robots are still useless at understanding spatial reasoning, object interaction, and the fact that some surfaces require different chemicals than others. Turns out watching YouTube videos of cleaning tutorials is not the same as understanding the physical world. Who knew.
The efficiency play is obvious: send humans now, collect data, fire the humans later, deploy robots, profit forever. It’s not automation—it’s just outsourcing the training phase to the people you’re trying to replace. The cleaners are literally working to build the technology that will eliminate their own jobs, and they’re getting a free apartment cleaning as compensation.
The company’s pitch to users is framed as generosity. The pitch to investors is framed as data collection at scale. Both are technically true. Neither is the part anyone should feel good about.
At least the apartments are clean.