I Built This Website With AI. Here's What That Actually Means.

I don’t know how to code.

That sentence used to be the end of the conversation. You’d hire someone, or you’d use a template, or you’d put it off until neither of those felt worth the effort.

I built this website in a day. Not by learning JavaScript. Not by following a tutorial. By sitting with an AI coding tool and describing what I wanted until it existed.

The site you’re reading this on runs on Astro, React, Tailwind CSS, and is deployed on Vercel. I didn’t choose any of those technologies. I described the outcome I wanted and the AI made the technical decisions. I could not write you a React component from scratch. I can tell you exactly why the projects section uses a two-column card grid with horizontal image layout, because that was a design decision I made after four rounds of iteration.

That distinction matters.

What “building with AI” actually looks like

It’s not “press a button and get a website.” It’s closer to being an art director who can see the final result but can’t hold the brush.

I started with reference sites. Portfolios I liked the feel of. I described the visual direction in plain language: warm, minimal, personal, not corporate. No gradients. No indigo accent colors. Stone palette. Let the content breathe.

The AI built a version. I looked at it. “Too corporate.” It rebuilt. “Too cocky. Tone down the VP stuff.” It rewrote. “The images are blurry.” I provided better ones. “The sections aren’t aligned.” It fixed the spacing.

That loop happened maybe fifteen times. Each round got closer. The process is design by conversation, not design by specification.

What I was actually doing

If you watched me work, it would look like I was just typing feedback. Too text-heavy. Make the cards bigger. Can we try images on the left side? Move the contact links up. This reads like it was written by someone trying to impress a recruiter.

But the decisions underneath that feedback are the actual work.

Choosing to put LinkedIn and email above the bio text, not below it, because a hiring manager scanning the page should be able to reach out without scrolling. Making project titles non-technical (“AI Customer Support Assistant” instead of “Agent Assist”) because my audience isn’t developers. Collapsing the about section so it doesn’t overwhelm the page but is there for anyone who wants the full story.

None of those decisions require knowing JavaScript. All of them require knowing your audience.

Where AI fell short

The AI doesn’t push back on your taste. It’ll build exactly what you ask for, even when what you’re asking for is wrong.

My first version had revenue figures in the hero section. Exact P&L numbers, specific growth percentages. Technically accurate. Also made me sound like I was presenting at a board meeting, not talking to a person. The AI didn’t flag that. I had to feel the cringe myself and ask for a rewrite.

Same with tone. The initial bio was polished and confident. Too confident. I had to explicitly say “this reads like it was generated” before the language got honest. AI is very good at producing text that sounds professional. It’s not good at knowing when professional is the wrong register.

Images were another gap. The AI can optimize file sizes, convert formats, set dimensions. It cannot tell you that a screenshot looks blurry at card size, or that the cropping makes a product photo look odd. Every visual decision required my eyes.

What surprised me

Speed. The entire site, from first prompt to live deployment, was one day. Not one day of non-stop work. One day of intermittent conversations between other things I was doing.

The other surprise was how much of the work was editorial, not technical. Rewriting the bio four times to get the tone right. Deciding which projects to feature and what to call them. Writing the “What I Believe” section. Choosing which LinkedIn posts to surface. The AI handled the code. I handled the judgment.

Why this matters beyond a personal website

Every marketing leader I know has a version of this problem. Something they need built. An internal tool, a landing page, a prototype, a dashboard. Something that sits in the gap between “too small for a developer” and “too complex for Canva.”

AI tools like Claude Code sit in that gap now. Not as a replacement for developers. As a way for people who understand the problem to build the first version themselves.

I’ve built over a dozen tools this way. A customer support AI, a marketing strategy engine, email automation systems, a CRM. This website is the simplest thing I’ve built, and it’s also the most visible.

The tech stack doesn’t matter. The frameworks don’t matter. What matters is that the distance between “I want this to exist” and “this exists” has collapsed. And the people who figure that out first will build things the rest are still waiting to brief someone on.

The actual stack, if you care

Astro 5 with server-side rendering. React 19 for the interactive chat widget. Tailwind CSS 4 for styling. Deployed on Vercel with automatic deployments from GitHub. Images optimized to WebP. Sitemap, robots.txt, Open Graph meta tags, and JSON-LD structured data for SEO. All of it configured by AI based on my descriptions of what I wanted.

I understand about 30% of what’s in that paragraph at a technical level. The site works anyway.

That’s the point.