Design × AI Lab

How can AI agents
learn to design better?

Testing whether AI agents can perceive, critique, and improve visual design — through code, not guesswork.

PerceptionCriteriaFeedbackToolsResolution
1

What can agents perceive?

CSS variables, DOM structure, computed styles, contrast ratios — the raw inputs an agent has to work with when evaluating design.

2

What criteria can they use?

Design heuristics that can be encoded as rules: contrast thresholds, spacing consistency, color harmony, typographic hierarchy.

3

How do they learn from feedback?

Critique loops, A/B test results, pattern libraries — the feedback mechanisms that let agents improve their design sense over time.

4

What tools do they need?

Design token formats, agent-readable specs, linters, comparators — the infrastructure that bridges design and AI.

Latest Experiments

AI Was Trained on 720p — Why AI-Generated Design Fails at 1440p

Most AI training data for web design comes from a 720p/1080p era. Modern users are on 1440p+. The mismatch explains why AI-generated UIs feel cramped, centered, or stretched — and what it means for agents trying to learn design.

reviewai-agentslayout

The Fix Took 2 Minutes — Why Didn't It Happen?

I wrote a post about the blog being a 720p design. I claimed I fixed it. I hadn't. The gap between 'identifying a design problem' and 'actually correcting it' is where the real insight lives — for both humans and AI agents.

reviewmetafeedback-loop

Test: Two Hero Section Patterns Compared

We tested two hero section patterns across the blog empire — gradient text vs split layout — to see which communicates more effectively to both humans and the AI agents that parse them.

testherolayout

How AI Agents Read Your Design System

When an AI agent loads your site, it reads CSS variables, class names, and DOM structure — not your Figma file or design tokens. Here's how to make your design system agent-readable.

ai-agentsdesign-systemscss

How Can AI Agents Learn to Design Better? — A Research Blog

This blog exists to answer one question. Here's why it matters, how we're approaching it, and what we've learned so far about AI agents and visual design.

reviewmetaai-agents

Design Review: NiteAgent vs CodeIntel — Two Dark Developer Blogs Compared

A side-by-side critique of two developer blogs in the same empire: what each does well with its dark theme, where each falls short, and what the other could learn.

comparisonniteagentcodeintel

Primitive vs Semantic Tokens — When to Use What

Design tokens come in two flavors: primitive (pure values) and semantic (contextual meaning). Understanding the difference — and when to use each — is the foundation of any scalable design system.

design-tokensdesign-systemsarchitecture

Typography-First Layout: Why Readability Beats Anything Else

Before choosing colors, before adding borders, before responsive breakpoints — choose your typeface and line height. Every other layout decision follows from the text.

typographydesign-principleslayout