The brief was a landing page for a regional pest control company expanding into a second city. Two-color brand, a hero, three service columns, a service-area map, social proof, and a quote form. Nothing exotic. We had four AI design tools open in four browser tabs. Same prompt, same brand assets, same logo file, same competitor reference screenshots dropped into each tool. We started a timer.
Three hours later, one of the four had something we could hand to the client. The other three were cut.
The contestants
Figma AI (the generative features built into the main app, not the older FigJam toys). v0 by Vercel, which generates React + Tailwind from a prompt. Galileo AI, which produces editable Figma files from natural language. Uizard, which has been doing this longer than most and was the elder of the four.
Two of these are aimed at designers. Two are aimed at "skip the designer." We wanted to see which approach worked when there was a real brief, real brand constraints, and a client with real opinions.
What each one produced
Figma AI gave us a reasonable wireframe in about four minutes. The hero composition was generic, but the structure was sound, and because the output was native Figma, we could grab it and work on it with our usual tools. It respected the brand colors when we fed in the hex values. The logo placeholder it generated looked like a different company's logo, which was fine because we were replacing it anyway.
v0 produced working React code in about ninety seconds. The visual design was the v0 house style — clean, modern, vaguely Vercel-shaped, with too much rounded-2xl. Components were real components: a proper Tailwind config, sensible spacing scale, accessible form labels. As code, it was the strongest output. As a design that looked like our client's brand, it was the weakest.
Galileo generated three "variations" in about six minutes. They were the prettiest at first glance, and also the most superficial. The hero used a stock photo of a generic services worker that had nothing to do with pest control. The "social proof" section contained reviews from companies that do not exist. The map block was a screenshot of Vienna.
Uizard produced something that looked like a 2019 design tutorial. The composition was fine. The aesthetic was eight years behind. We spent twenty minutes trying to push it forward and it kept reverting to the same bland card-and-shadow vocabulary. We closed the tab.
Where three of them broke
Galileo collapsed under feedback. The client liked one variation and asked for a swap from the stock photo to actual pest-control imagery, plus a column reorder. Galileo's "iterate" function regenerated the entire page from a modified prompt, which gave us a completely different layout, color treatment, and font. Whatever we'd just agreed on with the client was gone. We tried three times. Three different pages. We could not get from version 1 to version 1.1; we could only get to version 2, 3, and 4.
Uizard had the same problem in a different costume. Edits produced wholesale changes. No stable design that could be nudged.
v0 had a subtler failure. The code looked great, but when we tried to swap in the client's actual brand fonts, v0 had hallucinated a font import for "Sans Forgetica Pro," a font that does not exist. It also imported a phone-input component from a package that wasn't in the project. v0 is "demo-ready, not production-ready" — fantastic for stakeholder presentations, frustrating once you start trying to ship.
Why Figma AI shipped
The thing Figma AI got right was knowing its job. It didn't try to be the whole design. It generated a starting structure inside an environment we already work in, gave us editable layers, and got out of the way. When the client came back with feedback — "make the hero green, not white" — we didn't ask the AI. We just changed it. The AI's job was the first 20%. The remaining 80% was us, the way it always has been.
The other three tools all, in different ways, wanted to own the whole design. That was the problem. The moment a real client said "move that thing two inches to the left," the tools that owned the whole design fell over, because they had no concept of "modify this specific thing"; they only had "generate something new."
What this means for our process
We've added Figma AI to the standard kickoff workflow on landing-page projects. Not for finals — for getting from blank canvas to a starting wireframe in five minutes instead of forty. A designer still does the design. The AI compresses the staring-at-an-empty-artboard step.
We're going to keep watching v0. For internal tools and admin pages it's already useful. For client-facing brand work it's not there yet, and the shape of the problem suggests it may never get there without a different interaction model — something that can edit, not just regenerate.
Galileo and Uizard we're done with for now. Pretty first drafts are not what we needed. We needed editable second drafts, and neither tool understood the difference.
"AI design tool" is two different products pretending to be one. Some are trying to replace the designer. Some are trying to help the designer. The ones helping are the ones still in our tab bar this week.