Mimetic
Machines
Our Mission
Mimetic Machines is a data lab for physical intelligence.
Human skill is the largest body of knowledge never written down. It lives in hands: how a tailor judges tension in a seam, how an electrician feels a connector seat, how a cook moves without looking. That is why machines can write poetry and pass the bar exam but cannot fold a shirt.
Every intelligence we have built learned from a record. Language models had the written internet, centuries of it. Physical skill was never written down because it never needed to be: you learned it by standing next to someone who had it. That worked for ten thousand years. It does not work for a machine.
We record skilled human work as structured, verified data, and we build the benchmarks that measure whether a machine actually learned the skill. Imitation is the oldest form of teaching there is. René Girard spent a career showing how much of human behavior runs on it, and imitation is also, exactly, how a machine acquires a physical skill from a person. We think that is the shortest path to machines that are useful in the world of objects.
Team
We are engineers and researchers from the Technical University of Munich, CDTM, Stanford, and Berkeley. We have shipped AI products into some of the largest companies in Europe, built robotics and machine learning systems at Amazon, Google, BMW, and EGYM, and run operations across three continents.
We are working on what we believe is one of the most important problems of the decade: teaching machines to do physical work. The people who solve it will decide what the next century of labor looks like, who owns it, and who benefits from it. We would rather build the future we wish to see than wait to be handed someone else’s version of it.
We are based in Munich and we work where the work is.
Want to join the team
We hire high-agency operators with founder instincts and a quality bar that borders on perfectionism. We value high integrity, and building real value over perception.
You will thrive at Mimetic Machines if you are aggressive, competitive, and have a taste for changing and disrupting things.
We are a small team and we learn the way our models do. We find people who are excellent at something, we watch closely, and we copy what works.
We are hiring across operations, engineering, and hardware.
Head of Field Operations
You stand up capture sites from nothing: find the factory, win the floor manager, recruit and train the operators, and own throughput and quality once the cameras are rolling. This job is a passport, a spreadsheet, and a lot of time on real floors. You have run physical operations before and you know that the plan changes the moment you land.
Founding Data Engineer
You own the path from raw capture to a dataset a frontier lab will train on. Synchronization, derived signals, quality gates, packaging, delivery. Every episode we ship carries proof that the task was done right, and you build the machinery that makes that proof cheap. You have shipped data infrastructure that other people depended on, and you have opinions about formats.
Research Engineer, Robot Learning
You train policies on our data and you own the benchmarks that prove the data works. You will be the first person to know whether what we collect actually makes a machine better, and you will be trusted to say so out loud when it does not. Background in imitation learning, vision-language-action models, or robot manipulation.
Capture Systems Engineer
You design the rigs: cameras, sensors, calibration, and mounts that survive a full shift on a working floor and get handed to someone who has never used one. Cheap, repeatable, and rugged in that order. You have built hardware that left the bench and met the world.
Don’t see the right role? Pitch yourself anyway: careers@mimeticmachines.ai
Get in touch
If you are a researcher training or deploying embodied models, we probably have what you need (and so much more!). Write to team@mimeticmachines.ai
If you run a place where skilled work happens, we pay for time on camera and we handle consent end to end.