Writing/2026-05-18
A founder note on Semanttic, AI-generated UI, and why distribution has to shape the product from the beginning.
In November 2022, GPT-3.5 dropped and I had a holy cow moment.
That summer I had been quietly prototyping with earlier versions of ChatGPT, poking at what it could do with UI generation and front-end code. The results were interesting but not yet actionable, and I figured we were probably two or three years away from something real. Then 3.5 arrived and almost everything I had been trying to do that summer just worked. The tipping point had come earlier than expected.
Within weeks I had a working prototype that could generate a simple UI screen from a plain text prompt. At the time, that was genuinely new. We were among the first to demonstrate that capability publicly.
From there the insight felt obvious: if you could generate UI from a prompt, the next logical step was to connect that generation layer to real design systems, such as IBM Carbon, SAP Fiori, Microsoft Fluent, and Google Material.
Production-quality output, not just something that looked plausible in a demo.
We built that, ran enterprise pilots, and saw real results: faster delivery, dramatically lower production cost, the kind of numbers that make finance people pay attention.
We got pretty good at the design system integration. So good that Volvo's design team ran an audit comparing our output to one of their designers' mockups, and we were essentially dead on.
We went through Techstars, I was on This Week in Startups, where I got the first check from Jason Calacanis, established partnerships with Microsoft, Amazon, and Google, and eventually exited in April 2025.
The thing I would tell a founder building in this space today is this: figure out your distribution model first, then work backwards.
The question is not just "can we build something people want?" It is "how will people find this, how do they onboard, and what does it take to become the obvious choice before a larger player makes a competing feature announcement?"
Distribution and go-to-market are not things you figure out after product-market fit. They are part of the product.
For Semanttic, we never quite figured out the social media or viral distribution flywheel like Lovable and a couple others did, so they became the breakouts from our cohort. Meanwhile, the incumbents, like Figma, were mobilizing their armies of developers and designers to bring powerful new AI features into their existing products with millions of existing subscribers.
I am now applying that thinking inside Volvo Group, where the challenge is almost the inverse of a startup.
The technology is moving just as fast. The organization is not.
Manufacturing, supply chain, and purchasing are not spaces that generate viral demos or trend on Product Hunt. But they are enormous, full of deeply manual processes that have not changed in decades, and the leverage from compressing cycle times, even modestly, is significant.
The distribution problem in enterprise looks different: you do not need a flywheel. You need the right stakeholders to see something work once, in their context, on their data.
Getting there is its own discipline.