A private chef in a home kitchen now can be more than a luxury service. In Shift’s model, it can also be a data collection session for the next generation of home robots.
The arrangement is simple: a person receives a free or discounted service, while the worker records the task from a first-person viewpoint. For robot developers, that footage can capture the detailed hand movements, tool use and household context that machines will need if they are expected to work inside real homes.
Why a free lunch can be valuable data
Shift, a division of the German startup Microagi, sent a private chef named Ali to cook a three-course meal in an apartment. He arrived wearing a baseball hat with a blocky, white camera attached to the brim. The recording setup captured his work as he prepared the meal, including a gazpacho appetizer, trumpet mushrooms, baked salmon with a creamy zucchini sauce and tiramisu.
The point was not only to provide lunch. The footage was meant to become training data for robots, especially data showing skilled hand movement from the person doing the task. This kind of first-person recording is described as egocentric data collection.
That distinction matters. A cooking robot would not only need to know what a finished dish looks like. It would need to understand how hands move around knives, pans, ingredients, counters and sinks. A kitchen is full of small decisions and physical motions that are hard to capture through text alone.
Shift previously attracted attention by offering free cleaning services in New York City under a similar arrangement. The company later expanded in San Francisco by offering free home-cooked meals. In both cases, the exchange is the same: the household gets a service, and the company gets recordings of real work in a real home.
The marketplace behind robot training data
Microagi CEO Bercan Kilic frames Shift as more than a service promotion. “The goal of Shift is twofold. First, allow people to join the AI economy,” he says. “The other part is to handhold the entire society and governments together, from today until the abundance.”
In the near term, Kilic sees a marketplace where contractors can earn money by recording videos that help improve AI models. The source article also describes a broader gig-work pattern around physical movement data, including recordings made with an iPhone strapped to someone’s head, footage for DoorDash while scrambling eggs, and repeated recordings of fingers tying shoelaces for another data collection startup.
This is a different kind of AI labor than writing prompts or labeling images. The valuable asset is the body in motion: hands cooking, fingers tying, a person cleaning and navigating around household objects. For companies building robots, the home itself becomes part of the dataset.
The reason is straightforward. Large language models were powered by huge amounts of text scraped from the internet, but the source notes that there is no comparable vast repository of egocentric video data for robots. That absence helps explain why a company would trade a private chef or a cleaning service for recordings inside someone’s home.
What robots need to learn at home
Cooking shows why the challenge is bigger than following a recipe. Ali’s knife work, the handling of ingredients, the movement between counters and the cleanup afterward all created useful examples of physical skill. A home robot that cooks but leaves pans in the sink would still be incomplete, so cleaning data can be as important as cooking data.
The home also introduces mess, awkward layouts and personal spaces. A small apartment kitchen is not a controlled lab. That is precisely what makes the footage valuable: it reflects the conditions where a future humanoid robot would actually be expected to operate.
Shift’s model also changes the role of the household. The person receiving the service is not performing the task, but their environment becomes part of the recording. The worker’s hands are the primary subject, yet the kitchen, tools, surfaces and ordinary domestic context are all captured as part of the training session.
That can make the exchange feel unusual. Being recorded by technology in a private home is different from watching someone cook. The source compares the feeling to being around Meta’s AI glasses in public: the presence of a camera changes the social experience, even when the stated purpose is technical.
The promise and the risk of capable home robots
Kilic is optimistic about the speed of progress. “By this time next year we will have quite good home robots that are at reasonable prices,” he says. “It won't do all the tasks, but it will do most tasks quite well. It will be like vacuum robots in 2020.”
That vision depends on robots becoming useful with tools, food, surfaces and household routines. But giving AI models physical bodies also raises a practical safety concern. The source draws a contrast with a Roomba: a vacuum robot can make a serious mess when it fails, while a future chef robot using a knife could create a more direct danger if something went wrong.
There is also a broader social question. Kilic describes a possible future of abundance, but the source remains skeptical that robotics alone would eliminate poverty or solve wealth inequality. A robotics revolution could improve baseline living conditions while still leaving major economic problems intact.
For now, Shift’s free meals and cleaning services reveal the early shape of that future. Before home robots can cook, clean and move safely through private spaces, companies need examples of humans doing those things. That makes everyday domestic work, and the first-person video of it, a valuable resource in the race to build capable home robots.