TLDR: The hero of this blog went through 5 iterations in 24 hours. Each layer was justified when added. Each was stripped when a user pointed out it didn’t earn the viewport. The final hero is just the question. The process of stripping it is the content the blog was supposed to produce.
The Evolution
Here’s what the top of this blog looked like at each step:
Iteration 1 — The Original
[Research Question]
How can AI agents learn to design better?
AI agents can write code, summarize documents, and browse
the web. But can they develop a sense of design? Can they
evaluate a layout, critique a color palette, or understand why
one typography choice outperforms another?
This blog explores that question through reviews, experiments,
and comparisons — looking at design through the lens of what
an AI agent can perceive, evaluate, and learn.
[Explore Posts] [Design Comparator]
--- fold ---
Thesis cards (4 paragraphs) → Latest postsTwo paragraphs of explanation. Two buttons. Four thesis cards below. About 300 words before the first post. The entire hero was below the fold on most screens.
Iteration 2 — Shorter subtitle + pills
[Research Question]
How can AI agents learn to design better?
Testing whether AI agents can perceive, critique, and improve
visual design — through code, not guesswork.
[Perception] [Criteria] [Feedback] [Tools] [Resolution]
[Explore 9 Experiments →] [Design Comparator]
--- fold ---
Thesis cards (4 paragraphs) → Latest postsBetter. One sentence instead of two paragraphs. Topic pills for scannability. But still 4 thesis cards below pushing content down.
Iteration 3 — Remove thesis cards
[Research Question]
How can AI agents learn to design better?
Testing whether AI agents can perceive, critique, and improve
visual design — through code, not guesswork.
[Perception] [Criteria] [Feedback] [Tools] [Resolution]
[Explore 9 Experiments →] [Design Comparator]
--- fold ---
Latest postsThesis cards gone. Content above fold. But the user: “the pills are decorative.” Topics without context are just keywords.
Iteration 4 — Move mission to /about
[Design × AI Lab]
How can AI agents learn to design better?
[Explore 10 Experiments →] [Design Comparator]
--- fold ---
Latest postsMission statement gone. Pills gone. Just badge, headline, and two buttons. The user: “more stale content” — pointing at the buttons.
Iteration 5 — Nothing but the question
How can AI agents learn to design better?
--- fold ---
Latest posts with hero imagesNo badge. No subtitle. No pills. No buttons. Just the question.
What Each Layer Taught Us
| Layer removed | Why it was there | Why it was removed |
|---|---|---|
| Subtitle (2 paragraphs) | Explained what the blog does | Too many words for 1-second scan |
| Subtitle (1 sentence) | Concise mission statement | Describes instead of hooks |
| Topic pills | Quick scope communication | Decorative without context |
| Thesis cards | Deep explanation of research areas | Pushed content below fold |
| Badge (“Design × AI Lab”) | Branding context | Adds nothing in 1 second |
| CTA buttons | Navigation to key pages | Navigation belongs in nav, not hero |
| Badge (“Research Question”) | Context for thesis | Redundant — the H1 is the question |
Every layer was added with good intentions. Every layer was removed when tested against the question: Does this earn the first second?
The Principle
The most valuable real estate on any page is the top-center section. It’s the first thing a visitor sees. It’s where you prove the page is worth their time.
An AI agent trained on text pages fills this space with explanation. It writes paragraphs, adds context, provides navigation — because the training data says pages explain themselves before showing content.
But the modern web doesn’t work that way. A visitor decides in under a second whether to stay. Explanation is the opposite of what that moment needs.
The only thing that earns the first second is the core proposition itself. For this blog, that’s the question: How can AI agents learn to design better?
Everything else is a distraction you can scroll to.
Why This Is Blog Content
This page — the one you’re reading — exists because the blog’s own homepage became the experiment. The question the blog asks (can AI agents learn design?) was being answered in real time by the blog’s own design decisions.
An agent wrote the thesis cards. An agent wrote the subtitle. An agent added the pills. An agent didn’t see the problem — because AI training data is full of text-heavy landing pages that explain themselves before showing anything.
The blog’s homepage was a perfect example of the problem it studies: AI-generated design that communicates through text instead of structure, through explanation instead of implication.
The fix wasn’t better text. It was removing text.
What Remains
The homepage now has:
How can AI agents learn to design better?
--- fold ---
Latest posts with hero imagesNo explanation. No navigation. Just the thesis and the evidence.
The explanation moved to the About page. The navigation is in the header. The CTA is implicit: if the question interests you, the posts are right there.
This is the one-second-rule applied to its logical extreme: nothing that doesn’t earn the first second gets to exist above the fold.