Here is a fact about the current moment in software that is either exciting or alarming, depending on where you sit: a large language model will happily write you a login screen. It will pick a shade of blue, round the corners of a button, and space the form fields however it likes. The code compiles. The demo works. Everyone claps. And then a designer opens it, and the blue is the wrong blue, the button is the wrong button, and the spacing violates a rule that took the company three years and one very long Slack thread to agree on.
This is the problem Supernova has decided to make its business. The company, founded in Prague in 2018 by Jiri Trecak and Oskar Koristka, sells the unglamorous but load-bearing idea that AI can only build on-brand product if it actually knows the brand. Not the vibe of the brand - the specifics. The button variants. The spacing tokens. The color palette with its exact hex values and the reasons some of them exist.
“Product development is too slow and too complex, and it's holding back innovation.”
Jiri Trecak, Co-founder & CEOThe technical name for the thing Supernova sells is a design system: the shared library of tokens, components, documentation and rules that keeps a company's software looking like it came from one company rather than forty. For a long time, design systems were mostly a human coordination tool - a way for designers and engineers to argue less about handoff. Supernova's bet, sharpened considerably over the past year, is that the most important consumer of a design system is no longer a person at all. It's the AI coding agent sitting inside Cursor, GitHub Copilot or Claude Code.
So Supernova positions itself as the layer in between. It connects to design editors like Figma, reads the design system, and turns it into structured data - a single source of truth. Then it distributes that truth in formats machines can consume, including over MCP endpoints, the emerging plumbing that lets AI tools pull in context. The tagline the company now leads with is admirably blunt: “Design & engineering context, ready for AI agents.”
It is worth pausing on how contrarian this is. The dominant story about AI coding is that the models will keep getting smarter until context stops mattering. Supernova is wagering the opposite - that the models are already smart enough, and the bottleneck is that they don't know your button component. You do not fix that with a bigger model. You fix it with better structure. This is a less thrilling narrative than “the AI will do everything,” which is possibly why it is also more likely to be true.