The Two X’s Problem - by CodeYam and Dani

CodeYam’s Substack

SubscribeSign in

The Two X’s Problem<br>Why AI-designed brands feel like AI-designed brands

CodeYam and Dani<br>May 18, 2026

Share

When I was studying design, I made an album cover for the band The XX as a design exercise. I picked one of their songs, found a brush-painted font on dafont (those free falopa fonts sites that were all over the internet back then), typed out the title, called it a day.<br>My typography teacher looked at it and told me that he felt fooled. The two X’s were identical. Of course they were. They were the same glyph in a font file. But then, the whole point of brush lettering is that every stroke should be a little different…<br>I have been thinking about that a lot lately.

The XX’s album cover, circa 2011.<br>At CodeYam, I have been working on creating multiple design systems in Claude so they can work as starting points for our users to choose from for their projects. Themes I’ve created include Minimalistic, Gummy, Ethereal, Coder, and others.<br>The basic idea is to feed Claude Code (or the AI agent used within CodeYam) a mood board that gives the AI some direction in order to then get back a system the user can riff on. And it works. Fast. But my feelings about the initial results are mixed.<br>Sometimes, Claude nails it. Sometimes, while the result is technically correct, the output feels emotionally hollow. I have been reflecting on this and trying to understand what separates the two.

Some of the design systems Claude helped me create applied to dashboards: Minimalistic, Gummy, Ethereal, Coder.<br>What AI is genuinely great at

AI is great at creating generic artifacts. And I don’t mean that as an insult. Generic is what gets your MVP (Minimum Viable Product) out the door this week instead of the next month.<br>Even before AI, leveraging preexisting libraries, templates, and toolkits was a common practice to save time in design and development. Why spend time reinventing the wheel when there’s a decent enough set of components already there?<br>If you want a clean SaaS (Software-as-a-Service) landing page with a big Sans Serif headline, a soft gradient, and a CTA (call-to-action) button, Claude will deliver it before you finish your coffee.

Some examples of fintech, crypto, and dev tool websites.<br>Same with generic results for verticals like fintech (deep darks, blues, sharp grids), crypto (slightly weird, gradients pretending to be holographic), dev tools (mono accents, terminal greens). The internet is full of these patterns which have influenced the data used to train large language models (LLMs) that power AIs like Claude.<br>For example, I almost always see the same Google Fonts reused: Roboto, Open Sans, Lato, Inter, Montserrat, Poppins, Playfair, as well as their monospace cousins. Beautiful fonts, all of them. I love them. As a typography lover, I recognize how functional and well-made they are. But these fonts are everywhere now. There’s a reason every early-stage startup landing page is starting to look the same.

Some of the actual most popular fonts based on font views (source).<br>For an MVP, this is fine. Your site needs to be up and running fast.<br>Where it falls apart

But when you want something that stands out, that’s where everything falls apart. Instead of generic, you want unique. The weird stuff. The subtle stuff. The stuff with soul.<br>When I asked Claude for something more editorial, more brutalist, more illustrated, more specific to a feeling, the output got shaky. Misaligned elements. Strokes used in strange places.<br>Claude gave me a result that felt like the “almost there” version of something. If another designer had done it by hand, I would have thought that this was either a first step leading to something intentional and great – or a beginner’s mistake. From Claude, the results when asked for this kind of output typically land as a beginner’s mistake. And bad, lazy design.<br>There are a thousand quiet decisions that make a brand feel like it is saying something with value and in its own voice. Details such as the space between elements, the micro transitions, and the typography set with care instead of selected from a dropdown. Photographs with intention. Beautiful illustrations. Texture. Each of these make a difference when you’re trying to design a truly unique, and great, brand or product.<br>What’s actually missing

I keep going back to this: brand identity is not a logo and a font pairing. It’s a feeling, an intention, someone’s actual life and work. And conveying all this into a design system, comes from iteration, from understanding the competitors and the market, from knowing the mission of the product, from being willing to throw away three versions because none of them are right yet.<br>When we use AI to jump from idea to identity in one prompt, we skip exactly the part where the identity gets made. We get the artifacts of a brand without the process that produces one. And it shows.<br>This is the two X’s problem...