The company and the work it shows
Reimagine Robotics was founded in April 2025 by four people from Google DeepMind's applied robotics team: Jonathan Scholz, Oleg Sushkov, Akhil Raju and Misha Denil. Scholz founded that team and led it for seven years. The company is based in London and Sydney. It raised pre-seed financing from Fly Ventures and firstminute capital together with angel investors; the amount was not disclosed. No customers were named either.[1]
The arrangement works like this: the worker shows the robot the task, watches its attempt and corrects the mistake on the spot, which removes the need for a specialist programmer. The company describes two setups. At a made-to-order plastics facility a robot tends 3D printers; in a three-robot cell built for material recovery, hard drives are disassembled. The only efficiency quantity disclosed is that getting a task to a testable state fell from one day to roughly ten minutes.[1]
Where the gain is credited
The work that disappears has to be named correctly. What shortens in this arrangement is the step of programming a task into the robot; the demonstrating and correcting step does not go away, and it stays with the worker. The ten-minute figure therefore replaces time a programmer used to spend, and part of that time is transferred to the person on the floor. Scholz says as much: a robot that learns from the person doing the work turns that person's knowledge into leverage, and the company describes it as a tool that amplifies human labour. Whose hands the leverage stays in is absent from that account. Whether the behaviour produced by demonstration sits with the customer or the vendor, whether it can be transferred to other customers, and whether headcount at the two facilities changed are all unstated.[1]
The reasonable conclusion is that the ten-minute gain is credited first to whoever was buying the programming work: prototyping time is the line where a specialist's fee is measured. A weaker explanation also holds: the same fall could be obtained with nothing transferred from the person on the floor, simply by an existing engineer using better tools, and there is no working-hours data in the two setups described that would separate the cases. What divides the two explanations is two unpublished figures.[1]
The thing to ask plainly
The company set out plainly that it exists and where it works; the distribution question can have an answer of the same plainness. Headcount and contractor hours at the two facilities before and after installation, plus the contract term that says who keeps the behaviours produced by demonstration. Those three pieces of information are the shortest route to showing what the person who transfers their knowledge to the robot gets out of the transaction, and none of them has to be a trade secret. If those three are published, the distribution question stops being an argument and becomes arithmetic.[1]