Vibe coding for designers works, but not as type a prompt, get a product. How I go from Figma to a live build with Claude, and where it breaks.
Vibe coding for designers means describing what you want to an AI model in plain words and letting it write the code, so a designer can get a working app or website without a developer in the loop. It works, and I use it every week. But the version that ships something people want to touch is not "type a prompt, get a product". It is a real design first, a clear idea of how every piece should behave, and then an evening of small, boring fixes until the build matches what was in your head. The people who ask me for "the prompt" are usually skipping the part that makes it work. My honest view after building my experiments this way: vibe coding removes the handoff, not the design. If you skip the thinking, you get the same generic page everyone else gets.
This is how I actually do it, where it breaks, and when I would not use it at all.
Andrej Karpathy coined "vibe coding" in an X post in February 2025, describing a way of coding where you give in to the vibes and forget the code exists. By November 2025 Collins Dictionary had made it its Word of the Year for 2025, defined as using AI prompted by natural language to help write computer code.
That definition is written from the developer's side. For designers, the interesting part is different. For most of my career, my job ended at the handoff: polish the screens, write the specs, send the file, wait weeks to see something I could tap. That step is gone for me now. I still design every screen. I just do not stop there.
So when designers search "vibe coding for designers", I think the real question underneath is: can I ship my own work now, and will it be any good? Yes to the first. The second depends almost entirely on what you do before you open the chat.
I wrote the short version of this on X in a post called "I stopped handing off designs. I ship the product now". Here is the longer one.
I start by collecting, never in Figma. Dribbble and Behance for the feeling. Real products and Mobbin for how things should work, because somebody already solved the checkout flow. And AI images for the world the product lives in. When the world exists first, the UI has somewhere to live and it stops looking like a SaaS template.
I still open Figma first. It is the best place to think and get the layout right. But the file is not what I hand over anymore. It is the brief I give to Claude. Through the Figma MCP server, Claude reads my frames directly: layout, colors, fonts, images and effects. Two rules I do not break: frames at the real device size, and real effects (real blur, real shadows), so what I design is what ships.
I do not draw every state. For a loot box interaction, Figma held two frames: the closed box and the reward after it opens. Everything in between, the charging, the burst, the card spinning out, I describe in words. Drawing 40 in-between states would be the old me.
SwiftUI for apps, a live website for web. I talk to Claude the way I would talk to a developer friend, typos included. The water in my Pulse Health Liquid Glass experiment is not a video. It is a simulation running on the phone, and every glass you add pours in.
The first build takes minutes. Making it feel right takes the evening. On my Haribo pick and mix 3D experiment, I spent far too long because the bears were not falling right. That is the step most vibe coding guides skip, and it is where the craft is.
People see these builds and ask for the prompt. There isn't one. Here are real messages from my chats, word for word:
The bears aren't falling right.
Let me drag the card to spin it. Add sounds, and make the snow feel real.
None of those is clever. They only work because I already knew exactly what "right" looked and felt like. The AI is fast. The judgment is still mine.
The design can also change halfway through, and that is the best part. In Figma, the loot box said "swipe up to open". On my phone, swiping felt cheap. So I asked for hold instead: a glass ball appears under your finger, fills while you hold, dives into the box, and the card bursts out. In the handoff days, that idea would have died in a ticket. Now I feel it the same evening and change it.
And sometimes the AI is just wrong. A video model once made a sky look amazing and then quietly added raindrops to a lake on a sunny day. You only catch that if you stare at it like a designer. That part never got automated.
I do both, and I think the difference matters for how you grow. Vibe coding is a technique. Design engineering is a role: a designer who owns how the product is built and behaves in production. At ViewStats I worked in the real codebase with Cursor, Claude Code, React, TypeScript and GitHub, shipped components and opened pull requests. That is a different level of responsibility than a weekend build. The full story is in my ViewStats creator tools case study.
| Vibe coding | Design engineering | |
|---|---|---|
| What it is | A way to build by describing intent to an AI | A role that owns design and its implementation |
| Typical output | Experiments, landing pages, demos, small apps | Production features inside a team codebase |
| Who reviews the code | Often nobody | Engineers, through pull requests |
| Main risk | Polished surface, weak logic, hard to maintain | Slower, needs real code literacy |
| Where taste shows up | Motion, feel, the one memorable moment | Same, plus states, edge cases and system consistency |
| Best first step | Ship one small idea live with a link | Open one small pull request on a real product |
If the design engineering side is what you want, I wrote a practical path in how to become a design engineer, and the definition is in what a design engineer is in software.
I would be lying if I said it always works. These are the limits I hit or watch for:
Where I would not vibe code alone: anything that handles payments, health data or permissions without an engineer reviewing it. For an experiment, a pitch demo or a marketing moment, go for it.
The thing I would steal from my process, even if you never touch Claude: design the one moment people remember before anything else. Water pouring into a glass. An icy card you tap to shatter, like in my Payno shatter to unfreeze experiment. A film you scrub with your finger in the Traveme onboarding experiment.
When building was expensive, that moment was the first thing cut from the sprint. "Ship it static, add the animation later." Later never came. Now motion and sound are among the cheapest parts of the build, so they are the parts that make people stop scrolling. That is why my experiments have physics, motion and SFX, not just screens. You can try all of them in my AI design experiments.
Good, if you treat it as a way to ship your design, not a replacement for designing. It removes the wait between idea and something you can touch. It becomes bad when it replaces the thinking: then you get generic output that looks like everyone else's.
Not to start. You need to describe behavior precisely and notice when something feels off. Reading code helps a lot once you want to fix things yourself or work inside a team codebase, which is where design engineering begins.
The ones that respect your design file. My stack is Figma for the brief, Claude with the Figma MCP for the build, SwiftUI or a live website as output, and a video model for motion code cannot do. I compare tools side by side on my Claude Code vs Cursor comparison.
I do not think so. It replaced the handoff in my work, not the judgment. I wrote more on that in will AI replace product designers. The designers who will struggle are not the ones who cannot code. They are the ones still waiting for someone else to ship their work.
For me the first build takes minutes and the polish takes the evening. Things that used to take months take a couple of hours, but only when I know exactly what I want before I start.
A simple next step: pick one screen you designed this year that never shipped. Design its two key moments in Figma, write three sentences on how it should feel, and build it tonight. If you would rather have someone design and build it with you, work with me as your product designer.
If you have a web or mobile app idea in mind or you need some advice about product design, feel free to get in touch ✉️