AI for UI & Visual Design

I designed an infinite canvas in Figma, with product images floating freely across a white background, draggable in any direction, with hover cards revealing name and price. Then I gave the same brief and the same reference image to four AI coding tools and documented what each one produced.

same prompts, same references, different results↓

Closest Replica (3 prompts)

On Claude Code

Farthest Replica (7 prompts)

On v0.dev

I'm learning to work with AI by pushing it until it breaks, and paying attention to what happens.

My prompting strategy

Layout before anything else

Nothing else matters if the spatial structure and interactivity is wrong. I'd try to get the canvas positioning, density, breathing room right before fixing colors and fonts.


Isolate one component, be specific

Vague feedback gives vague results. I made sure to be so detailed in my prompts, that there is no room for the AI tool to do guesswork.

Cursor (4 prompts)

faithfulness—————— 75%

deviation——————— incorrect layout spacing

surprise———————- breathing animation on images

v0 (7 prompts)

faithfulness—————— 30%

deviation——————— redesigned identity

surprise———————- useful microcopy

Figma Make (7 prompts)

faithfulness—————— 60%

deviation——————— images stacked, not scattered

surprise———————- unprompted micro-interactions

Claude Code (3 prompts)

faithfulness—————— 90%

deviation——————— increased kerning in the title

surprise———————- almost identical to my reference

I'm learning to work with AI by making a note of what works well, and what doesn't.

The first result is decent at best. Patience make something good something great.

The first result is decent at best. Patience make something good something great.

The higher fidelity your reference, the less the AI has to guess.

The higher fidelity your reference, the less the AI has to guess.

Some AI suggestions are great. Knowing which ones to keep and which ones to override requires judgment.

Some AI suggestions are great. Knowing which ones to keep and which ones to override requires judgment.

I'm learning to work with AI by understanding both of our weaknesses and strengths.

What AI is good at

  1. Handling the technical execution

  1. Iterating quickly once direction is clear

  1. Generating a working base fast

What I bring

  1. Design eye to evaluate output vs. intent

  1. Precise prompting strategy

  1. The design vision

What I'm still learning

What I'm still learning

What I'm still learning

#1

Knowing which tool to reach for before starting

a project.

#2

Reading the AI code output, not just the visual output.

#3

Understanding the full capabilities of AI as a design tool.

I'm learning to work with AI
one experiment at a time.

Copyright 2026