The bottleneck in industrial automation has not been the robot for a long time. Arms are cheap, reliable, and available on lead times measured in weeks. The bottleneck is that every time the task changes, somebody with a specific and expensive skill has to come and reprogram the thing — which is why automation lives in factories that make the same object a million times and stays out of the ones that don’t.
On Monday, August 3, a London and Sydney startup called Reimagine Robotics came out of stealth with a product aimed squarely at that bottleneck, and one number that deserves attention: during a customer deployment, it cut the time to prototype and test a new robot behaviour from roughly one day to about ten minutes.
The founders are not a random garage team
Jonathan Scholz founded Google DeepMind’s Applied Robotics team in London and ran it for seven years. He co-founded Reimagine in April 2025 with three former colleagues — Oleg Sushkov, Akhil Raju, and Misha Denil. The pre-seed round, amount undisclosed, came from Fly Ventures, firstminute capital, and angels.
Scholz calls the approach “monkey-see, monkey-do.” A worker shows the robot a task, watches it try, corrects it when it gets it wrong, and moves on. No specialist required in the loop.
“A robot should arrive with the attitude of a new colleague: ‘How can I help? What do you want me to do?’” he said. “The people who understand the process should be able to answer those questions by showing the robot directly.”
He is emphatic that this is augmentation, not replacement. “For us, this is not about taking people out of the process. A robot that learns on the job depends on people. The worker identifies the bottleneck, shows the robot how to help, and corrects it until it is useful.” He describes it as “a tool to amplify human labour.”
Take that at face value. It is very probably what he believes, and in the near term it is probably true. It is also not the part that matters most.
Ten minutes is the number that changes who gets automated
Read the two deployments Reimagine disclosed. At a made-to-order plastics business, the robots were trained to tend 3D printers overnight — removing print beds, working latches, pressing controls — and then the customer’s own team used the platform to automate washing, curing, and drying without calling anyone. In a separate project recovering critical materials from used hard drives, the company built a three-robot disassembly cell alongside process engineers.
Made-to-order plastics. Hard-drive teardown. These are precisely the jobs that were previously uneconomic to automate, because the task varies and the reprogramming cost dominated. Drop the iteration cost from a day to ten minutes and the arithmetic inverts. A shop that could never justify a robotics integrator can now justify a robot, because retraining it is a Tuesday afternoon rather than a purchase order.
Note also what got automated first at the plastics customer: the overnight shift. Machine tending, at night, alone. That is not a coincidence of the pilot — it is the highest-value, lowest-resistance slice in any small factory, and it is a shift somebody used to work.
Amplification and displacement are the same mechanism at different speeds
Scholz’s framing and the displacement reading are not in conflict; they are the same fact described at two time horizons.
Phase one is genuinely amplification. The worker is the training signal. Their judgement about what’s worth automating and their correction of the robot’s mistakes are the product — you cannot run this system without them, and it makes them more valuable, not less. That is real.
Phase two is what happens once the demonstrations accumulate. Reimagine is on a list the industry knows well — 1X, AgiBot, Apptronik, NEURA, Sanctuary AI, Tutor Intelligence — all of them using human operators to generate training data for robot foundation models. The strategic logic of that entire category is that human demonstration is the scarce input now, and the goal is to need less of it later. Every correction is both a fix and a datapoint.
None of which makes Reimagine’s pitch dishonest. It makes the offer to the worker clear: you are the expert, your knowledge is what the system needs, and the value you provide is being recorded. “Instead of someone having to repeat a tedious physical task thousands of times, they can teach the robot, and apply that ability wherever it is needed,” Scholz said. That is a good trade if “wherever it is needed” includes you. The next fundraise will say a lot about whether it does.
