Last weekend, zutato.com went live. Not the product yet – that comes in autumn. But the site where you can sign up for the beta. And in parallel, we filed for the trademark.
Two sentences that read harmlessly. There is more history behind both than the website itself suggests.
What Zutato is
Zutato is a food-tracking app that doesn't stop at three macronutrients. Up to 138 nutrients per product, broken down into eleven categories, with transparent sourcing. If you want to know where a value comes from, you can see it – manufacturer label, calculation, estimate, or user entry.
The claim behind it is simple and uncompromising: quality over quantity. Rather a smaller, carefully maintained database than a few million unreviewed entries with half of them wrong. Every product is reviewed by the team before it goes public. Sounds like extra work – and it is. But that's the only way to build a database you can actually trust when it comes to micronutrients.
Beta launch is autumn 2026. Right now, you can sign up at zutato.com, nothing more – and that's enough for now.
Ten years, several attempts
The idea behind Zutato has been with me for almost a decade. The footer doesn't say "© 2014 – 2026" by accident. In that time, there have been multiple attempts, multiple prototypes, multiple technical dead ends.
There was a phase where I went deep into OCR experiments – extracting nutrition tables from photo scans, long before that became a standard problem you could explain to a modern AI in two lines. There were phases where I tinkered with the data foundation, with sync strategies, with architectural decisions that turned out months later to have been the wrong questions.
And there was the most painful lesson: a good idea on its own doesn't build a good product. What does is the right understanding of what it's meant to be – and what it isn't.
That took time. Longer than I would have liked. But in hindsight, every one of those attempts makes sense. Without the dead ends, today's version wouldn't be what it is.
Why now
What has changed in recent months: the technology is at a level where the product can work without investing five years of lead time into OCR pipelines, image processing, and manual data curation. AI takes a large part of that load away – not as a replacement for human review, but as a pre-filter. Photo in, values out, human checks.
That's the point at which the product became feasible without needing a ten-person team.
Right now we're two. The domain knowledge for what's in the product comes from the team. I build it.
The trademark
In parallel to launching the website, I filed for Zutato as a word mark. That sounds like a footnote. For me, it was new ground.
It was my first trademark filing, and I genuinely underestimated the process. Which classes do I need? How narrow or broad should I phrase the list of goods and services? What does "software" mean in which class, what falls under "providing information"? Which fees are due when? What do I do if someone files an opposition?
A small odyssey. Nothing dramatic – but at three different points, I asked myself whether I was filling out the right form or whether I would receive a rejection in six weeks because I had confused class 9 with class 42.
The application is out now. The next few months will tell whether everything fits.
A Saturday evening for the website
The website itself was done in one Saturday evening.
Concretely: Nuxt 4, Tailwind, the usual toolchain. Written with Claude Code as pair programmer – the same way I built my own portfolio site. That wasn't a coincidence, it was a deliberate choice: known stack, known workflow, known setup. No experiments on an evening when I want results.
What does "one Saturday evening" mean in practice? A concept in my head for weeks, a rough outline in notes, a clear sense of tone and color palette. Then sat down, went through the site section by section with Claude Code, wrote the content in parallel – and by the end of the evening, the site was up. Beta form included, GDPR compliant, imprint, privacy policy.
This doesn't work because "AI takes over everything." It works because three things come together:
- Clarity beforehand. I knew what I was building. No structural questions on the evening itself.
- Familiar tools. Not a single new stack component. Every command landed.
- AI as accelerator, not architect. Claude Code writes faster than I can type – but I decide what gets written.
Building a product website on a Saturday evening doesn't work via vibes. It works because the specification was in my head first.
What I take from this
The idea took ten years. The website took an evening. That isn't a contradiction – it's exactly the point.
The hardest part of a product isn't the implementation. It's figuring out what it's actually meant to be. Which problems it solves, which it deliberately doesn't, who it's for, and why. That clarity can't be accelerated. Not even with AI. It needs iteration, dead ends, frustration, the occasional thought about giving up.
What can be accelerated is everything that comes afterwards. Once the idea is sharp enough, clarity translates into speed.
An idea takes a decade, a website takes an evening. Anyone who mixes those up is either building too fast – or not at all.
Originally published on grossbyte.io
Top comments (0)