How I Built My Personal Website With One Prompt and Claude's Fable 5

TL;DR: I built jigarrana.com, the site you are reading right now, in one working session with Claude’s Fable 5 model. I wrote one detailed prompt to my Cowork setup, and the system handled the rest: design exploration, a written spec, an implementation plan, the code, and the deployment. My job was making decisions, and the total came to about a dozen approvals from first prompt to live domain.

The old way of using AI was the problem

Most people use AI tools the same way: type a request, get something close, correct it, get something closer, correct it again. Twenty prompts later you have a result and a headache. I spent months working this way before I understood that the problem was how I was briefing the tool.

The fix was giving the system everything it needed up front. One prompt with the full context: what I wanted to build, who it was for, a reference site I admired, and the constraint that it should be clean and minimal. From there, the AI ran the project and I made the calls.

What one prompt actually produced

My prompt was a few sentences. I pointed to a website I liked as inspiration, said I wanted a personal brand site for recruiters and LinkedIn traffic on my own domain, and asked for help picking the design direction.

From that single brief, the system:

Ran a design consultation. It mocked up three visual directions in my browser, with live color palettes and mini homepage previews. I clicked the one I liked. It offered accent color alternatives when I said the blue felt too much like LinkedIn, and we landed on an oxblood red pulled from the pocket square in my photo.
Wrote a spec. Positioning, page structure, design system, technical architecture, and hard rules about my writing voice, all documented before any code existed.
Planned the build. A task-by-task implementation plan with the actual copy, written in my voice, included in the plan itself.
Built and verified. An Astro static site with perfect Lighthouse accessibility and SEO scores, checked against my banned-phrase list by a script it wrote for the purpose.
Deployed it. Live preview on Vercel, then my domain connected, with me pasting two DNS records into GoDaddy as my only manual step.

The photo on the homepage went through the same pipeline: the system removed the background using a small Swift script it wrote on the spot, cropped it, and placed it in the hero.

The stack, for those who want specifics

Model: Claude Fable 5, running in a Claude Code workspace. Framework: Astro 5, a static site generator that outputs plain HTML, which keeps the site fast and easy for search engines and AI crawlers to read. Hosting: Vercel’s free tier. Design: Anton and Instrument Sans, two free fonts, with one accent color. Total cost: the domain I already owned.

Why this matters if you don’t write code

I am not a developer. My background is in business. What changed this year is that the gap between having an idea and having a working product collapsed, and the skill that matters now is briefing: knowing what you want, providing the context, and making decisions quickly when options come back.

That skill transfers from my day job. A good prompt looks a lot like a good brief to an agency or a good ticket to an engineering team.

Frequently asked questions

How long did it take to build? One working session from first prompt to live site, including design decisions, copy review, and DNS setup.

Did it really take one prompt? One prompt started the project and carried it through design, planning, and build. After that, my inputs were decisions and refinements: picking a design direction, approving copy, requesting changes. Deciding and re-prompting are different things, and the difference is where the time savings live.

What does a setup like this cost? A Claude subscription, a domain name, and free hosting on Vercel.

Can AI-built websites rank on Google and show up in AI search? Yes, when they ship the fundamentals: static HTML, semantic structure, fast load times, schema markup, a sitemap, and an llms.txt file for AI crawlers. This site includes all of them, and they were part of the build rather than an afterthought.

Jigar Rana is a tech enthusiast and AI-first operator based in Toronto. If you want to talk about AI-first ways of working, book a call or connect on LinkedIn.