How I Rebuilt My Website in 5 Hours with AI
The exact process, prompts, and decisions I made to go from a generic page to a professional site in a single day using Claude Code.
The Problem
My previous page was a Next.js template with generic text that didn't represent anything I actually do. The typical landing page you make "for now" that stays there for months.
What I needed was clear: a page that showed real projects with real numbers. That felt like a developer portfolio, not another agency with stock photos.
The problem is that designing a good website takes time. Hiring someone takes more. And I had a window of a single day between classes and work.
What I wanted to achieve
- •A dark, minimalist site, Linear/Vercel style — that communicated "builder", not "agency"
- •Real projects with metrics (2,500+ students, 89K payments, 24/7 agents)
- •Direct copy — no filler words, no AI slop
- •Something that could be built in a weekend
If your website doesn't reflect what you do today, it's hurting you. Better to have no website than one that gives the wrong impression.
The Initial Prompt
I didn't start by telling Claude Code "make me a nice website." That gives generic results. Instead, I passed my complete brand manual (colors, fonts, tone of voice) and asked it to interview me.
Here's my brand guide with colors, fonts, and tone.
Before building anything, ask me questions about what I want
to communicate, who my audience is, and what projects to highlight.
Interview me as if you were a professional web designer.
Claude Code asked me about 15 questions: what type of visitor I expect, whether I want a portfolio or landing page style, what feeling I want to give, what things I don't want to appear. That conversation was more useful than any brief I've ever written.
Why it works
AI can't read your mind. If you give vague instructions, you get vague results. But if you let it ask questions first, it extracts information you didn't even know you had clear. The interview replaces the discovery process that normally takes days.
Saying "make me a professional and modern website" is the worst prompt possible. You'll get something generic that looks the same as the other 10,000 websites the AI generated with the same instruction.
Research Agents
After the interview came the first version of the site. The structure was good, but the copy was terrible. Generic. "Cutting-edge solutions" and that kind of garbage that says nothing.
So I ran research agents: I asked Claude to analyze how the best tech pages present their projects, what copy they use, how they organize information.
Research how the best tech pages present case studies
(Linear, Vercel, Raycast). Analyze:
- How they describe their products in one line
- What metrics they use and how they present them
- The tone — what they avoid saying
- The visual structure of each project/feature
Then apply those patterns to my 6 projects.
What I discovered
- 1The best pages describe WHAT the product does, not HOW they built it. The tech stack goes separately, as tags. The description sells the result.
- 2Each project carries ONE concrete metric. Not three, not five. One. "2,500+ students managed" says more than a paragraph.
- 3The copy avoids superlatives. No "revolutionary" or "cutting-edge." The numbers speak for themselves.
When AI gives you generic copy, don't tell it "make it better." Give it concrete examples of what you like and ask it to analyze the pattern before rewriting.
The 5 Hours
Everything happened in a single day. 5 hours of real work, with breaks in between. Here's how it went:
- Hour 0-1Interview with Claude Code. Defined audience, tone, structure. Passed the brand guide (colors, fonts, voice rules).
- Hour 1-2First complete version of the site: hero, projects, about, contact. Solid structure, generic copy.
- Hour 2-3Ran research agents to analyze copy from the best tech pages. Rewrote all project descriptions with the new framework.
- Hour 3-4Visual iterations: spacing, subtle animations, responsive, refined dark mode. At one point I left it working alone and went to watch TV.
- Hour 4-5Final adjustments, deploy to Vercel, mobile testing. Site live.
What I learned
Most of the time wasn't "building" — it was directing. AI generates fast, but making good decisions about what to keep and what to change is the real work. Architect, don't code.
Want to see the result?
The site is live. Go check it out and tell me what you think.
Made with Claude Code · 2026