Artur Grzeda

13 years on the web: front-end, no-code, now agent systems. Building @_devlooper at @wearetonik, a Figma-to-code tool that grades its own work.

[Artur Grzeda @_ArturGrzeda]

made a text dial for swiper. words on a wheel, a ring, a fan, a 3d barrel or a straight rule - the live one reads straight, the rest turn away and fade.

and it clicks. eight detent sounds, synthesised in code, no audio files 🙌

[Artur Grzeda @_ArturGrzeda]

280 presets → https://t.co/leHvaRjoSE

[Artur Grzeda @_ArturGrzeda]

made a text dial for swiper. words on a wheel, a ring, a fan, a 3d barrel or a straight rule - the live one reads straight, the rest turn away and fade.

and it clicks. eight detent sounds, synthesised in code, no audio files 🙌

[Devlooper @_devlooper]

100 on the waitlist.

If you're seeing this for the first time: devlooper turns a Figma file into production code. You point it at the design at the end of the day and walk away. Overnight it builds the page, scores every section against what you drew, and keeps fixing until it matches. In the morning you have a repo you own and a deployed link.

Thank you.

https://t.co/uD44RrpmGB

[Artur Grzeda @_ArturGrzeda]

Designed by you. Built without you.
https://t.co/RJggO5NNGA

[Artur Grzeda @_ArturGrzeda]

orbs orbs orbs ! https://t.co/5xinTXPbnu

[Artur Grzeda @_ArturGrzeda]

saw @alex_barashkov's carousel distortion and wanted it as something reusable, so I made it a swiper plugin. one shader pass, 150 presets 🙌

[Artur Grzeda @_ArturGrzeda]

want to try? https://t.co/LJ0JxSCy4K

[Artur Grzeda @_ArturGrzeda]

saw @alex_barashkov's carousel distortion and wanted it as something reusable, so I made it a swiper plugin. one shader pass, 150 presets 🙌

[Artur Grzeda @_ArturGrzeda]

Designed by you. Built without you.
https://t.co/RJggO5NNGA

[Artur Grzeda @_ArturGrzeda]

orbs orbs orbs ! https://t.co/5xinTXPbnu

[Artur Grzeda @_ArturGrzeda]

@_devlooper runs for hours: building, judging, fixing. I wanted the UI to show which one, not a spinner meaning "something, somewhere".

found nothing worth using, so I built it.

want it as a package? like + comment and I'll open it up.

https://t.co/VVYKFsZiYv

[Artur Grzeda @_ArturGrzeda]

@webflow The link to the article doesn't work for me!

[Artur Grzeda @_ArturGrzeda]

@_devlooper runs for hours: building, judging, fixing. I wanted the UI to show which one, not a spinner meaning "something, somewhere".

found nothing worth using, so I built it.

want it as a package? like + comment and I'll open it up.

https://t.co/VVYKFsZiYv

[tonik - design for startups @wearetonik]

We got tired of choosing between shipping fast and owning the code. So our own @_ArturGrzeda built Devlooper: one agent turns your Figma into code, a second grades it against the design until it passes. We've been shipping our own client work with it. Now we're letting people in. Waitlist now open 👇 https://t.co/DKkh37UVys

[brryant @bryantchou]

The toughest brand review we ever did wasn’t with our agency. It was with my 6-year-old.

Here’s why every startup needs to care more about brand right now (and exactly how we built ours at ploy):

AI is making every product look the same.

Same fonts. Same gradients. Same “clean” SaaS vibes.

The only real way to break through the noise anymore
is a visual brand that people actually feel.

When I started ploy, I followed five simple rules:

The name had to mean something
The visuals had to be recognizable to kids
It needed to blur art&humanities + tech
It had to feel fun
The brand itself should get stronger as AI image models improve

Then I just… did the work.

Every morning for two weeks I hunted domains on InstantDomainSearch.
Day 14: in my dreams I wondered if “ploy” was available.

turn out it was. bought it in under 60 seconds for a lot of dough.

Next I talked to our design partners at tonik (the best in the game) and said:
“I love Jeff Koons’ balloon dogs. I’ve never seen anything like that in startup branding. Don’t copy him… but make something that feels familiar and completely new.”

They went completely nutso.
Shapes, textures, and animations that actually feel alive.

Then came the real test.
I showed the balloon motifs to my kids.
My 8-year-old rolled her eyes.
My 6-year-old looked at them and said… “ok.”
If you’re a parent, you already know: that’s a standing ovation.

Finally I tried generating the assets with AI. nano banana 2 got the weird shapes surprisingly close.
The animations still needed a real Blender artist.

Lesson?
Brand is no longer optional.
In a world where everything looks the same,
the companies that feel different win.

Look around the arts, nature, Finnish, for inspo.

Make something a 6-year-old would approve of.

[Artur Grzeda @_ArturGrzeda]

Figma file in. real site out. checked against the design at every breakpoint. you start it and walk away

roughly 80% of the page. the rest is still my job.

I spent a few months on this tool. built at @wearetonik

[Devlooper @_devlooper]

I built a tool that turns a Figma file into production code. The part I'm proudest of is what it won't do. It never invents the design.

That wasn't laziness. A brand isn't a logo or a palette. It's the hundred choices a designer makes and the thousand they rule out, and that's the whole distance between a site that's unmistakable and one that's interchangeable.

A model can't work that way. A model averages. It hands you the middle of everything it has ever seen, which is why the pages it invents all arrive the same: gradient hero, three icon cards, a stat bar nobody asked for, centred CTA. Your client pays to stand apart and receives their competitor's website.

You cannot average your way to a decision.

So devlooper starts where your file ends. Point it at the design when you finish for the day, and go home. Overnight it builds, scores every section against what you drew, and keeps fixing until it clears the bar. By morning: production code in your repo, and a link.

The score measures fidelity, not quality. It will confirm the build matches your design. It will never tell you the design was any good.

That part is still yours. I've yet to see anyone take it.

What it does take is the pettier job. Opening staging next to Figma. Logging the heading that's one weight heavy, the section twelve pixels off, and the three more you delete because they sound small. All caught now, before the link reaches you.

Everything downstream of your file is becoming a commodity.

Your file isn't.

https://t.co/uD44RrpUw9

[Devlooper @_devlooper]

Introducing devlooper

The whole path from a design to a deployed site

The first full version of a page comes out of one run, overnight, with nobody in the room.

Point it at a Figma file and walk away. It builds the page section by section in Next or Astro, straight into your repo, grades every section against the design at every breakpoint, and keeps fixing until it clears the bar. Roughly 80% of the page is done before you look at it. Not one output handed over and called done. A loop that stops when the work matches.

→ A whole site, not one page. List the pages and it builds them, each held to the same bar
→ Every section graded against the design at every breakpoint, and fixed until it passes
→ The pages nobody designed, a privacy policy or a blog index, you describe in a sentence and it builds them to match the rest
→ Then you take over. Pick any element on the live page, give notes like you would to a designer, it edits the real source
→ Commit, CI gate, deploy to Vercel, without leaving the app

It built its own landing page last week: 18 rounds, 113 minutes, 86 against a bar of 85. It still needed a person for the last 20%, which is the part that needs taste.

No API keys. It runs on the Claude plan you already have, and your files never leave your Mac.

Built at @wearetonik, the studio behind the Supabase, LangChain and LlamaIndex front ends.

Designed by you. Built without you.

Waitlist: https://t.co/uD44RrpmGB