Why I'm Building My Personal Site Entirely With AI
An experiment in using Claude Code and AI tools to scaffold, design, and ship a production website — and what it means for agencies.
BY MARCELO RUSSO
This site you're reading was built entirely by AI tools. Not "AI-assisted" in the way people usually mean — where a human writes most of it and ChatGPT fills in some copy. I mean the full stack: scaffold, architecture, content, AEO optimization, deployment. All of it.
Why Bother?
Two reasons. First, I genuinely wanted to see if current AI tools (specifically Claude Code) could produce something I'd actually ship. Not a demo. Not a prototype. A real site that represents me professionally.
Second, I run an agency that's increasingly powered by AI. If I'm going to tell clients that AI can produce production-quality work, I should be able to point to my own site as evidence.
What Worked
The technical scaffold was fast and correct. Next.js 15, App Router, TypeScript, Tailwind, structured data, AEO optimization — all generated in a single session. The AI understood the architecture decisions and made good defaults.
The structured data was actually better than what I'd typically write by hand. The Person schema, Organization schema, FAQ schema, Article schema — all properly linked with @id references. Most agencies skip this entirely or do it poorly.
What Required Human Judgment
Content voice. AI can write competent copy, but matching a specific person's voice requires iteration. I had to guide it toward my actual way of speaking — direct, slightly irreverent, occasionally mixing in Spanish.
Design decisions. The AI produced clean, functional design. But "clean and functional" is also "safe and forgettable." The interesting design choices — the ones that make a site feel like it belongs to a specific person — still need a human pushing for them.
What This Means for Agencies
AI isn't replacing agencies. It's changing what agencies spend their time on. The commodity parts of web development — scaffolding, boilerplate, structured data, basic content — those are now nearly free. What's left is the hard stuff: strategy, voice, design judgment, client relationships.
That's a good trade. The busy work was never the valuable part anyway.
Frequently Asked Questions About Building Websites With AI
Can you build a production website entirely with AI?
Yes. This site — themarcerusso.com — was built entirely with AI tools including Claude Code, from scaffold to deployment. The full stack was AI-generated: Next.js 15 architecture, TypeScript, Tailwind CSS styling, structured data markup, AEO optimization, and Vercel deployment. The key distinction is that AI handles the technical execution while a human provides strategic direction, voice, and design judgment.
What AI tools are best for building websites?
Claude Code is currently the most capable tool for full-stack website generation, handling everything from project scaffolding to component architecture to deployment configuration. For design iteration, tools like Midjourney or Figma AI assist with visual direction. The most effective workflow combines an AI coding agent for technical execution with human oversight for brand voice, design taste, and strategic decisions. No single tool replaces the full process yet, but Claude Code comes closest for development.
How long does it take to build a site with AI?
A production-ready personal site or marketing site can be scaffolded, built, and deployed in a single working session — roughly 4 to 8 hours — compared to 2 to 4 weeks with traditional development. The time savings come primarily from eliminating boilerplate: routing, component structure, structured data, meta tags, and responsive layouts are generated in minutes rather than hours. The human time shifts toward content strategy, voice refinement, and design direction.
What are the limitations of AI-generated websites?
AI produces technically correct but aesthetically safe designs. The output tends toward "clean and functional" rather than distinctive or memorable. Brand voice requires significant iteration — AI can write competent copy but matching a specific person's tone takes guidance. Complex interactive features, custom animations, and novel UX patterns still need human direction. The technology is best suited for content-driven sites, marketing pages, and structured applications rather than highly experimental or artistic projects.
Should agencies use AI to build client websites?
Agencies should use AI for the commodity parts of web development — scaffolding, boilerplate, structured data, basic content generation — and redirect human time toward strategy, client relationships, and design judgment. At Fri3nds Agency, AI handles roughly 60% of the technical execution on new builds, which lets the team focus on the work that actually differentiates the output. The agencies that resist AI tooling will lose on speed and price; the ones that adopt it thoughtfully will deliver better work faster.