I Asked AI to Pick My Brand Colors. It Asked Me Four Questions Instead.

I am not a designer. I have a PhD in marketing and AI, two years as a CTO, and zero intuition for color theory. When I decided to build a personal website for my freelance AI consulting practice, I knew the visual identity, the personal brand, would be the part where I needed the most help.

I had just finished building a brainstorming plugin for Claude Code. The plugin exists because LLMs have systematic failure modes during creative work: premature convergence, sycophancy, mode collapse. Each of its eight methods is a countermeasure for a specific, named bias. I had tested it on technical problems. Now I wanted to test it on something personal.

The question seemed simple enough: what colors should my brand use?

#The Brief I Gave Myself

I started by writing a research plan. Four parallel agents, each assigned a different lane: competitor visual identity analysis, modern AI/tech brand inspiration, color psychology and palette design, typography and design system research. The plan was detailed. The assumptions baked into it were not.

I knew what I wanted. Or thought I did.

“Modern / cutting-edge. Professional / trustworthy. Fresh / breath of fresh air, not another corporate navy-blue consultant. Elegant / simple. Dynamic / sporty.”

I had a reference site I admired, one I remembered as having a dark, neon aesthetic. I was leaning toward neon yellow on dark grey. The plan told the agents to research neon accent palettes on dark bases. The brief was confident.

It was also wrong about almost everything.

#Three Agents, One Surprise

The research agents fanned out. One screenshotted competitor sites via Playwright while the others ran in parallel: web searches for inspiration sites, color psychology research, typography analysis. About six minutes later, everything came back. 5 palettes with hex codes. 15 inspiration sites. 10 fonts analyzed. 4 typography pairings.

The most interesting finding had nothing to do with the proposals themselves.

The site I admired, the one I had mentally filed under “dark neon AI aesthetic,” actually uses a warm light palette. Deep plum and coral accents on a cream background. Not dark at all. Not neon at all. My reference point for the entire brief was a memory that didn’t match reality. The agent caught this because it actually visited the site and looked at it. I had been going on vibes.

The five initial palettes were concrete, each with hex codes, accessibility analysis, and mood descriptions:

“Volt” — electric yellow #E4FF1A on near-black #0A0A0A. The system’s top pick. Maximum differentiation.
“Neural” — electric cyan #00E5FF on midnight blue. The safe AI/tech default.
“Clarity” — electric blue #2563EB on white. Clean, contrarian, corporate.
“Forge” — copper #D4845A and gold #C9A96E on warm navy #0B1120. Bridges the old brand.
“Signal” — acid green #BFFF00 on warm white. The wild card.

Click through them. That is what the research produced. Bold, distinctive options. And, as it turned out, not what I wanted at all.

#The URL Detective

I wanted to see the palettes on a real page layout, not just as hex swatches. I pointed the agent at realtimecolors.com and asked it to figure out how the tool works. It navigated the site via Playwright, clicked through the font panel, ran JavaScript in the browser console to check whether settings persisted in the URL, and mapped the full schema: five hex values (text, background, primary, secondary, accent), two Google Font names, and a confirmed finding that type scale doesn’t persist. Within minutes I had working preview links for every palette on three different page templates. Also useful: Coolors for side-by-side swatch views.

#One Palette Out of Five

My girlfriend and I clicked through all five palettes. Only one resonated. Not the system’s top pick. Not the neon. Palette 4: “Forge.” Warm navy #0B1120 background, copper #D4845A accent, gold #C9A96E secondary, cream #F0EDE8 text. The warmest, most human option of the five.

She also made a point that stuck: blue is overused. Every second tech consultant, every AI startup landing page. Deep blue plus white plus a safe accent. We had both noticed it independently, but hearing it said out loud made it impossible to unsee. I had written “not another corporate navy-blue” in my own brief, and yet three of the five palettes the system proposed were variations on exactly that.

I asked Claude to brainstorm from a different angle. What could fit my personality? What directions had I not even considered?

Then I noticed something. The brainstorming skill ran inline, working through its methods sequentially in the same conversation. That is not what it is supposed to do. The whole point of the heavy tier is parallel subagents: fresh contexts, no anchoring from prior conversation, no sycophantic drift toward ideas already on the table.

“I thought the skill forces you to do that with subagents?”

It didn’t, in that version. Claude acknowledged the miss. (That catch later led me to change the plugin’s default behavior.)

“Redo it with the heavy tier.”

#Four Agents, Four Questions

Four subagents launched in parallel. Each pair used a different cognitive lens: First Principles paired with Constraint Manipulation. Inversion paired with Perspective Forcing. Analogy Search paired with MECE Decomposition. Assumption Surfacing paired with Diverge-then-Converge. They explored independently, each in a fresh context with no knowledge of the prior conversation. About six minutes later, they came back.

The most unexpected finding was not a color recommendation.

Three of four agents, working independently without seeing each other’s output, converged on the same surprise.

The warm colors I liked were already there. They were just trapped inside a dark background that made the site look like everyone else’s.

The agent running Assumption Surfacing put it most directly: flip to light mode, keep the warmth, and the palette becomes the most distinctive thing in the AI consulting space. Not because the colors change. Because you can finally see them.

The Inversion agent flagged the slide deck problem: dark navy turns to mud on projectors. Copper becomes indistinguishable from default presentation templates. Any palette that doesn’t have a strong light-mode variant has a practical failure mode baked in.

But the real output was not palettes. It was four questions.

Why copper? Aesthetic preference or conceptual anchor?
What does mastery look like to you? Visible and legible, dense and dark, or durable like patina?
Are you the trusted expert or the current expert? The agents drew a line I hadn’t seen before. On one side: the trusted expert, the kind signaled by established consulting firms and luxury watchmakers. “I’ve been doing this a long time. Trust the track record.” On the other side: the current expert. The kind signaled by modern developer tool companies. “I understand what matters right now. I’m building with the tools you’ll be using next quarter.” Both are premium. Completely different signal.
Do you present on projectors? Because dark backgrounds fail there.

I had not thought about my brand through any of these lenses. Especially that third question. I had been treating “premium” as one thing. The agents split it in two and asked me to pick a side. The system I built to pick colors had instead asked me who I was.

#The Current Expert. Clearly.

I answered the four questions honestly.

Copper was a design preference. I am drawn to warmth and elegance. I’m open to changing it. I’m not a designer.

Visible, legible, unpretentious. Light mode.

The current expert. Clearly. I’m not selling heritage. I’m selling the fact that I understand what matters right now. Production AI systems, current workflows, 2026 tools. Everything about my positioning is present-tense.

Yes, presentations matter. Online and at conferences.

Those answers eliminated every dark palette. Forge, the one I had been drawn to, reads as the trusted expert: heritage, craft, earned-over-decades authority. My proof points say the opposite. The palette was fighting its own positioning.

Then came the moment that surprised me about myself.

“I just said elegant because I like it. I am drawn to it. This maybe the complete wrong choice for my professional brand though. I am open to change.”

That sentence, typed into a chat window at 10 AM on a Friday, was the most honest thing I said in the entire session. I had built a brainstorming system that was now questioning my aesthetics, and instead of defending my taste, I let the questions land.

The agent built three variations on a light base called Praxis: warm white #FAFAF8, anthracite text #1E1B18, one disciplined accent color. It opened them in the browser via realtimecolors.com. Ember #A0401E. Near-monochrome #3A3A3A. Forest green #2D6A4F.

I rejected monochrome in under a minute. “Why did we come up with mono. It feels completely wrong to me. Even lifeless.” That reaction taught me something real: I am a colorful person. I like saturation and hue. Monochrome might be elegant in the abstract, but it is not me.

The session ended with several accent candidates on the Praxis base, warm and cool, each with preview links. No final choice was made.

Here is what I locked in: light mode. The current expert. Presentations matter. Copper is negotiable. Monochrome is dead.

Here is what I hadn’t decided: the accent color. The one that would make the brand mine. Look around — you’re seeing the answer right now.

#What the Plugin Actually Did

The brainstorming plugin is part of fractional-cto, a Claude Code plugin collection. I built it to counteract how LLMs predictably fail at creative work. I expected it to give me better color options.

It gave me better questions instead. Four questions I could not have asked myself, surfaced by four independent agents that had never seen my prior conversation. The color palette will follow. The self-knowledge was the harder part.