loading...
Enigma raised $71M to put 100+ robots online and watch strangers control them – the experiment is the product, and the data it generates is the moat.
ENTRY ANGLES
Run a narrowly scoped public interaction experiment for one specific robot category to generate a vertically specific dataset · Build the interaction layer for one industrial domain – warehouse, surgical, or agricultural robots – before Enigma's general model ships · Apply the same data-generation methodology to non-robot domains: voice interfaces for industrial machinery, or operator-correction systems for computer vision
VERTICALS
CAPABILITIES
Robotics hardware access, large-scale data collection infrastructure, ML/RL research, human-computer interaction design
The robots can already do more than anyone knows how to ask of them.
Jonathan Jacobi and Gal Niv – who met as teenagers competing in hacking contests and later served together in Israel's Unit 8200 intelligence unit – looked at the current state of robotics and arrived at a thesis that cuts against the dominant industry narrative: the technical capabilities of modern robots already exceed what operators can harness. The bottleneck is not the AI. It is the interface.
Enigma, less than one year old and funded at $71M by Index Ventures and Ribbit Capital, emerged from stealth on July 27, 2026 to test this thesis at scale. Rather than deploying robots into a market niche, the company has deployed more than 100 proprietary AI robots in hangars in Israel and California and opened the controls to anyone in the world. The robots can draw with paintbrushes, duel each other with swords, mix reagents in chemistry flasks. Every person who logs on and tries to control one is generating data on how humans naturally want to communicate intent to a machine.
The experiment runs across four interaction modes: text commands, audio instructions, video demonstrations of an intended motion, and direct manual control. Each mode tells the research team something different. Where do people default to text? When do they switch to audio? What kind of motion do they try to demonstrate when they can't find the words? How precise do they expect the response to be? The patterns across millions of interactions will shape both the eventual interface and the architecture of the foundation model that sits behind it.
Jacobi was Microsoft's youngest-ever employee before working alongside Wiz founder Assaf Rappaport. The seed round attracted researchers from OpenAI, Anthropic, xAI, and Cognition as angels – a signal that the people closest to the frontier of AI capability believe the interface problem is the real frontier for robotics.
The graphical user interface did not make computers more capable. It made existing capability accessible to people who hadn't wanted to learn command lines. The smartphone camera is not technically superior to a professional DSLR; it is accessible enough that billions of people use it for things professional cameras never reached. In both cases, the advance was not in the underlying system – it was in the layer between the system and the human.
Robotics is in the same position. Manipulation dexterity, environmental perception, and multi-step planning in commercially available robots already exceed what most operators know how to request. The gap between what modern robots can do and what operators can successfully instruct is not a model problem. It is a communication problem.
Enigma's experiment generates data on that communication problem at a scale no laboratory could match. The company is not crowd-sourcing robot control to find operators – it is using the public's interaction patterns to discover what "natural" robot communication actually looks like. The resulting dataset is proprietary to Enigma and non-replicable by any competitor that didn't run the same experiment.
Index Ventures investing alongside Ribbit Capital is an unusual pairing: Index covers software and marketplaces; Ribbit covers fintech. Their joint participation reflects a shared conviction that this is primarily a software and data problem, not a hardware problem – a categorization that matters for how the eventual business scales.
The global robotics market is projected to reach $218 billion by 2030. The fraction of that market currently limited by interface quality rather than mechanical capability is a large and defensible target.
Enigma is building a data collection and model training business that will eventually sell its output as an API: a foundation model for human-robot communication, licensable by any manufacturer or enterprise that wants its robots to accept instruction from non-expert users.
The commercial logic is clean. Enigma generates proprietary interaction data that no competitor has, because no competitor ran the experiment. That data trains foundation models. Those models are licensed to robot manufacturers, enterprise buyers, and software platforms building on top of robotics infrastructure. The experiment itself is the moat.
The same methodology applies outside robotics to any system that needs to accept natural human input in non-standardized form: voice interfaces for complex industrial machinery, computer-vision systems corrected by non-technical operators, enterprise software where the gap between capability and usability is the primary adoption barrier. The underlying research question – what is the most natural way for a human to express intent to a machine? – is not specific to robots.
Before Enigma generates its full dataset, there is room for narrower versions of this experiment. Pick one robot category – warehouse arm, delivery drone, surgical assistant – design a minimal public interaction experiment, and generate a dataset specific to how operators in that environment naturally communicate. A vertically specific dataset is worth more to that industry than a general one and defensible against Enigma's eventual general foundation model.