I Can't Draw, So I Hired a Robot Design Team: Making the HOME-LAB.net Logo

I can design a multi-region network. I can untangle a TerraformTerraformHashiCorp TerraformDevOpsAn open-source Infrastructure as Code (IaC) tool that provisions, manages, and versions multi-cloud resources declaratively via HashiCorp Configuration Language. state file at 2 a.m. What I cannot do is draw a logo.

Every time I’ve tried, I end up in the same place: a monospace font, a little >_ prompt, and a quiet voice saying “this is fine, developers love this.” It is not fine. It looks like every other side project on the internet.

So when I started planning home-lab.net, the home for my homelab, I wanted something better. Not a corporate mark, just a small, honest logo that felt like a sibling to the one on this site, and looked good whether it was tiny in a browser tab or sitting at the top of a page.

I can’t draw. But I can react. I know instantly when something looks right and when it looks wrong. So I tried an experiment: let an AIAIArtificial IntelligenceAIComputer systems and neural architectures capable of reasoning, code synthesis, pattern recognition, and autonomous agentic execution. design assistant throw ideas at me, and I’d play art director with nothing but gut feelings and short sentences.

It took twelve rounds, 23 mood boards, and 49 little vector files. Here’s how it went.

---

Round one: more server, more server, MORE SERVER

My first instinct was literal. It’s a homelab, so the logo should look like a server, right?

So the early sketches were gloriously over-the-top: tiny drive bays, rack ears with little screw holes, honeycomb air vents, network ports, even a miniature status screen. I asked for “more obviously server-like” and the assistant happily obliged.

It was impressive for about thirty seconds.

Then I imagined those logos at the size they’d actually appear, a thin bar at the top of a web page, and everything turned to visual noise. All those lovingly detailed drive bays shrank into a smudge. A logo has to survive being small, and mine couldn’t.

The lesson I kept relearning: take things away

Here’s the thing I didn’t expect. Almost every good decision in this project was a subtraction.

I’d look at a board of eight ideas and feel nothing. Then one of them would make me lean forward. It wasn’t the fanciest. It was the simplest: three plain white bars stacked on top of each other, each with a little slot on the left, and one tiny glowing blue light on the top one.

That’s it. Three bars. But somehow it said “server” and “stack of things quietly doing their job” and “someone cares about tidy cable management,” all at once.

From there, every round was me saying some version of “now remove something”:

• “Lose the rounded square behind it.” The bars looked better floating on their own.
• “Make the bottom bar sit on the line.” The underline under the name now starts after the bottom bar, with a small gap. It looks like the bar is holding the whole line up.
• “Make .net blue.” The little light on the top bar and the .net now match, a quiet wink between the two ends of the logo.

None of these were technical decisions. They were just me squinting and going “hmm, less.”

The part where the robot got it wrong

My favorite moment came near the end. I noticed the three bars were a touch too tall. The top one poked above the letters of “HOME-LAB” like a cowlick, and it was driving me quietly insane.

I asked the assistant to line them up. It dutifully produced a lovely diagram with a neat dashed line labelled “aligned”… and the bar was still sticking up above the letters. The picture-making side of the AI understood what an alignment diagram is supposed to look like, but it couldn’t actually line two things up.

So we switched tactics. Instead of asking for a picture of the logo, the assistant simply wrote the logo as a set of exact measurements, and everything snapped into place: the top of the first bar level with the top of the letters, the bottom bar sitting right on the underline, three bars with equal gaps in between. Done.

That was the real takeaway for me. The AI’s imagination is great for finding ideas, but you want something precise to build the final thing. Brainstorm with one, finish with the other.

The finished logo

Here it is, on a dark background and a light one:

Three bars, one blue light, one line. It works at the size of a browser tab and at the size of a poster. I could not have drawn it, and I could not have described it at the start. I only knew it when I saw it.

What I’d tell anyone else trying this

1. You don’t need to be a designer, you need taste. Knowing “not that, closer, now remove something” is a real skill, and you probably already have it.
2. Ask for lots of options, then pick by feel. Eight bad ideas on one page make the one good idea jump out. That’s cheaper than agonizing over a single attempt.
3. When the logo has to be precise, use precision. Let the AI dream up ideas, then have it build the winner carefully.
4. Keep subtracting. If you’re unsure what to do next, remove one thing and see if you miss it.

The logo is going on home-lab.net once the site is up. In the meantime, I have a very tidy folder of 49 logos that didn’t make it, and I’m weirdly fond of every one of them.