Why I Rebuilt My Website with Claude Code
My personal site lived on Wix for years. It worked. Six months of building with AI made it obvious I didn't need it anymore. Here's why I rebuilt in Claude Code — and what becomes possible once your site is actually yours.
For years my personal site lived on Wix.
It worked. I paid the bill. I logged in when I needed to change something. Every year or two I'd think "I should really update this," and every time I thought about what the actual rebuild would cost — time, developer coordination, design revisions, deployment — I'd close the tab and go do something else.
Wix wasn't the problem. What I could do with a site was the problem. And for a long time, I accepted those limits because the alternative was expensive.
Then I spent six months building with AI. And at some point in there, the math on a rebuild quietly flipped.
The six months that changed what was possible
If you've been reading this series, you've seen the journey — the 2am nights, the products shipped, the thousand-plus hours of vibe coding. By the time this past winter rolled around, I wasn't the same person who had put that Wix site up. I could build things in a weekend that would have taken a development team a full sprint. I could ship a production app in four weeks that would have been a twelve-month project under the old model.
And my own site was still a drag-and-drop template.
The gap between what I was telling people about AI and what my own personal site looked like had stopped being comfortable. If I'm spending my nights building AI products and writing about what's possible, my own site has to show the work. Fine is not how I want to represent myself. Especially not when better than fine had become a weekend.
So I rebuilt it. From scratch. In Claude Code.
The stack
I didn't overthink the decisions.
- Next.js App Router — the current default. Server Components where they help, client components where they're needed, file-based routing that stays out of the way.
- Tailwind CSS — fast to ship, easy to maintain, zero time lost on CSS preferences.
- MDX for writing — every post on this site, including this one, is a Markdown file with React components sprinkled in. No CMS, no database, no login. Git is my editor.
- Vercel for hosting — push to GitHub, it deploys. Preview URLs per branch. Analytics built in. Free at my scale.
- Claude Code as my pair.
Five decisions. One afternoon. The rest was building.
What Claude Code actually feels like to work with
I built it with Claude Code is the kind of line that can sound like marketing copy, so let me be specific about what a working session actually looks like.
I open Claude Code in the project. I describe what I want. Sometimes it's small — "the venture card has a bug where images don't load at small viewports." Sometimes it's bigger — "let's migrate all my portfolio images from my old Wix CDN and build a masonry gallery page for them." Sometimes it's architectural — "let's rename the /ventures route to /lab and add proper 301 redirects so we don't lose any inbound links."
Claude proposes an approach, lays out the specific changes, and makes them. Before I accept anything, I read the diff. Sometimes I push back — "use MDX for the post loader instead of a database," "don't use a carousel there, use a masonry grid," "that gradient is gross, try something cleaner." That pushback is what makes the sessions productive. It's the vibe coding thesis from earlier in this series, in practice: you have to have earned the vocabulary to direct the tool well. Otherwise you're just accepting whatever comes out.
When you know what you want and you can recognize when you haven't gotten it, Claude Code becomes a force multiplier. When you don't, it's a very fast way to ship something mediocre.
Real moments from the rebuild
A few specific examples so this doesn't stay abstract.
Pulling 45 photos off Wix's CDN. My old portfolio lived on Wix's image pipeline — compressed, cropped, transformed. I wanted the originals. Claude and I figured out the right URL pattern, wrote a script to pull all 45 full-resolution files, split them into a portfolio set and a behind-the-scenes set, and rebuilt the gallery page around them. Start to finish: maybe twenty minutes. Under the old workflow — email a developer, wait for a reply, review, iterate — that's a week. Minimum.
Renaming a whole section of the site. Halfway through building this series I decided /lab read better than /ventures for the Innovation Lab section. That's a small-sounding change that's actually annoying in any traditional CMS — rename the page, update every internal link, and make sure nobody's old bookmarks 404. Ten minutes with Claude Code: renamed the directory, updated every reference across five files, added two permanent redirects in next.config.ts, verified end-to-end. Done.
The blog pipeline itself. Every post in this series was drafted from a Q&A — I'd ask you questions, you'd answer, I'd transcribe the source material into a markdown file, then draft the post from that source. The Q&A files live in a _research/ folder inside content/writing/. They're not imported by the site, but they're there as receipts: every claim in a post can be traced to something you actually said. I wouldn't have set that process up in a traditional CMS. You can't. Because the writing is code, I can set up any process I want around it.
The Mike Aron Visuals gallery card. You wanted a richer image for the Alfred card on the Innovation Lab page. I scraped a fresh hero from hey-alfred.app, cropped it to the right aspect ratio, dropped it in, verified it was rendering, and moved on. Four minutes. No Photoshop, no asset request, no back-and-forth.
None of these are technically impressive on their own. What's impressive is the compound effect. The site has had roughly a hundred changes like these over the course of this rebuild. Every one of them would have been an email under the old workflow, and a lot of them simply wouldn't have happened.
What this makes possible next
The most interesting thing about building a site this way isn't what I've already shipped. It's what I can ship next without a meeting.
The clearest one I've been thinking about: an Ask Mike embedded chat. A real AI assistant, trained on my writing, my public perspectives, and my thinking patterns, embedded directly into this site so someone can just ask it a question. "What's your take on enterprise AI adoption?" "What should I read first if I want to build my first AI product?" "Walk me through your thinking on multi-agent architecture." That used to be a product in itself. Now it's a weekend project I could plug into this site the next time I had a quiet Saturday.
Same with new content types. Interactive demos. Live examples from the products on the Innovation Lab page. A newsletter pipeline. Video content. Each of those was a feature request in the old model, now it's a prompt in the new one.
That's the real unlock. Not that I rebuilt the site. That I unblocked myself for everything that comes after.
What this looks like for everyone else
If you're sitting on a Wix or Squarespace or WordPress site, wondering if it's worth rebuilding: the answer depends on one variable.
Are you willing to learn enough to direct the tool?
If yes, rebuild. It's worth it. You'll recoup the time within a month because every future change gets cheaper, every new feature becomes a conversation instead of a project, and the site itself becomes a place you actually want to keep improving.
If no, stay where you are. Vibe coding with no foundation is a trap. You'll ship something, it'll look okay, and then the first time anything breaks you'll be stuck. The cost isn't the rebuild. It's the maintenance after.
For me, six months of nightly AI work was the foundation that made this rebuild cheap. For someone who hasn't put in that time yet, the honest play is to build that foundation first. Pick a small side project. Ship it badly. Ship the next one. By the third one, you'll know whether you can direct the tool. Then come back and rebuild the site.
The site you're on right now
If you're reading this, you're on the result. Next.js. MDX. Tailwind. Vercel. Claude Code as the co-pilot through the whole rebuild.
Zero dollars in hosting. Full control over every pixel. A writing pipeline I actually enjoy using. An architecture I understand and can evolve. A surface I can add features to on demand.
No more Wix. No more CMS. No more "fine."
Just a site that's actually mine, built the way everything I'm building now gets built — with AI as a collaborator, not as a feature slapped on top.