My Frustrating Attempt to Outsource Creativity to AI

When I decided to create this business, I needed a name, a brand identity, a visual feel for the website. I don’t consider myself to be super artistic. That’s not to say I am not creative. I believe many of the things I do in software and AI require creativity, but picking colors and fonts and branding is not something I am particularly confident in. So I put off these decisions and instead developed what the words on the page would be.

When I code using AI, I treat it much more as a conversational partner, with a back and forth refining each other’s ideas. Not knowing where to start with these artistic decisions however, I decided to approach using AI as a mathematical function¹. I gave it the entire text of the website and asked it to come up with some ideas for names and color palettes, hoping for straightforward progress to make up for my lack of experience.

It was immediately clear that would not be the case. Multiple sessions, multiple days, multiple different LLMs, Claude, ChatGPT, Gemini, all led to the same average output. The same fire and light metaphors, the same fonts and colors, even the same exact words, Attune, Align, suggesting Signal, not factoring in that branding relies on unique identities.

I tried to improve the quality of the output. I tried asking it to give me more names. I tried telling it what I didn’t like about the options, as much as I could. I tried new sessions, with slightly different positioning of my asks. Nothing was leading to the results I wanted. Until I randomly got lucky one time.

One of my many sessions popped out the word docent. The quiet presence in the museum gallery there to help deepen your own knowledge and accompany you along your day temporarily. It was perfect, but of course there are already a bunch of docents in the AI space, so it was dead as a brand. But it was enough. I wanted a job, a role, to name my business.

Once I had that key, the frustration was gone. I wasn’t asking the AI to do my work for me, I was asking it to help me tease out why docent felt so right, and what else could capture that same energy, just as I do when coding. Very quickly Attaché appeared. The assistant in a diplomatic mission responsible for a niche area of expertise. I immediately bought the domain and here we are.

While all this was going on, I was also trying to build a visual identity and website for what would become Attaché. My first instructions to AI were woefully underdefined, and despite knowing and writing instructions for a site that didn’t follow the common patterns of an AI-generated website, I still ended up with one that had clearly regressed to the mean. It even used Space Grotesk, the exact font Anthropic themselves identified as a hallmark of AI-generated websites. No matter how many times I tried to use the built-in web search tools to expand both my and the LLM’s context at the same time, the mediocrity remained.

So I decided to take another approach: ask it to help me think through the process, the thing I should have done from the beginning.

It immediately asked what artists I liked. I don’t strongly identify with any particular artists, but I had recently been considering buying some bold posters for my house, so I decided to simply flip through some online poster stores. The posters of the JPL era, the new national parks series, certain ones from Olympic games and F1 grands prix really caught my attention. I gave them to the AI and we quickly established a poster-based language for the color palette and design.

This whole experience just reinforced my belief in using AI tools as collaborators, not as servants. I already knew this and had internalized it when I was working on things I was comfortable with, but completely let my unfamiliarity guide me to treating AI as a cure-all, not as a tool, and it led to days of frustration. They are really good at spitting out average on their own. But to make the output something you are happy with, you first need to give it enough information about what you are doing. And while traditional advice focuses on building the perfect prompt to do this all in one go, it is less overwhelming to ask to have a conversation about what you are trying to achieve, letting the goals and important pieces emerge naturally.²