On Design: Why Most AI Products Look the Same and What We Did Instead

There is a specific visual language that has emerged for AI products in the last three years.

Dark backgrounds. Purple gradients. Glowing orbs. Sans-serif fonts at sizes that feel important without saying anything. Animated statistics that load dramatically. Floating cards that demonstrate the product's output in a way that looks polished but doesn't quite show you how to get there yourself.

I think this language has become a liability.

The convergence problem

When every product in a category looks the same, the design stops communicating differentiation and starts communicating category membership. You're not saying "we are different from our competitors." You're saying "we are a legitimate member of this category." That's a floor, not a ceiling.

For a new entrant with no brand equity and no existing user base, looking like your competitors is actively harmful. The visitor arrives, sees a dark gradient hero with a purple CTA, and their brain files you under "AI tool — will evaluate later." You've spent design effort to be forgettable.

We had a choice: join the aesthetic consensus or reject it.

What Jony Ive actually said

I kept coming back to something from Ive's work at Apple. The principle wasn't minimalism for its own sake — it was the idea that every element in a design should earn its presence. If you can remove it and the design is better or unchanged, you should remove it.

The purple gradient orb floating in the background of every AI landing page — what is it doing? It's signaling "sophisticated" and "premium" without delivering either. It's decorative chrome that costs visual attention without paying back in comprehension or feeling.

We kept the dark background because it's right for the product and the audience. Streamers work in dark rooms. The content they create — gaming, late-night streams — lives in that aesthetic register. But we stripped everything else that wasn't earning its place.

The result: no decorative gradients on brand elements. No floating orbs in hero sections. No statistics that exist to impress rather than inform. The gradient is reserved for CTAs only, because there it does a real job — it signals "this is the thing you press."

The 18px decision

We set the base font size to 18px. This is larger than most web products default to (16px is the browser default, and many products go smaller for density).

The reasoning: at 100% zoom on a high-DPI display — which is how most of our target users are viewing content — 16px body text is slightly too small to read comfortably for longer than a few seconds. Our users aren't skimming legal documents. They're evaluating whether to trust us with their content workflow. We want them to actually read what we wrote.

18px says: we think what we wrote is worth reading.

The typography hierarchy

The POMS wordmark uses no icon. No sparkles, no lightning bolt, no abstract shape that's supposed to evoke "clips" or "speed." The name stands alone. This was a deliberate choice.

Icons on wordmarks are almost always a hedge. They exist because the designer wasn't confident the name was strong enough on its own. POMS is strong enough. The .gg extension carries enough context. Adding an icon would dilute both.

The .gg appears at reduced opacity — white/40 against white/100 for POMS. Not hidden, not absent, but clearly secondary. The name is the brand. The TLD is the context.

The comparison grid problem

Every AI tool has a comparison grid. A table with checkmarks showing that the product has everything competitors have, plus more. I've seen maybe two hundred of these grids across two hundred SaaS landing pages.

None of them are credible. Everyone knows the company filled in their own column. The bias is so obvious that the grids have become noise.

We built comparison pages — 16 of them, one per major competitor. But instead of checkmark tables, we wrote actual comparisons. Where competitors are better than us, we said so. Where we're better, we explained why with specifics. The goal was to write the comparison a trusted friend would write if you asked them to research your options.

Trusted friends don't give you 47 green checkmarks and 0 red ones.

What design is for

Design is not decoration. It is not differentiation theater. It is the structure that allows ideas to be understood.

A product that is difficult to understand — visually, structurally, rhetorically — is a product that is leaking conversion at every stage. Every element of friction between "visitor arrives" and "visitor understands what this does and why they should care" is a design failure.

We are trying to build something great. Not impressive. Great. The distinction matters. Impressive things make you say "wow." Great things make you say "yes."

Everything in the design is oriented toward "yes."