Building with Claude AI
I asked claude to generate this story. It needs several adjustements for it to be Me, change focus, and describe what i found really important myself. But not bad for a starter.
I didn’t build this website alone. Claude AI was part of the process from the start — and working with it taught me things I didn’t expect.
My background is Oracle and PL/SQL. Frontend work was always at the edges for me: enough to get by, not enough to feel fluent. When I decided to build this site in SvelteKit, I knew I’d be learning as I went. Claude became the pair programmer I didn’t have.
What it actually looked like
The workflow was conversational. I’d describe what I wanted — a two-column links page, a favicon with VLC as a monogram, a green glow on hover — and Claude would write the code, explain the reasoning, and adjust when something didn’t look right.
What surprised me was how much I learned in the process. Because Claude explains its choices, not just delivers code, I started understanding why things work. Why the static adapter needs prerender = true. Why SVG favicons built from paths are more reliable than text-based ones. Why MDsveX is a better fit for content pages than plain Svelte.
I wasn’t just accepting output. I was asking follow-up questions, pushing back, making decisions. Claude handled the implementation; I kept the direction.
Where it helped most
A few areas where having Claude alongside made a real difference:
- Getting started fast — No blank-page paralysis. Describe the structure, get a working scaffold.
- Debugging edge cases — Small things like prerendering configuration or favicon caching in Edge that would have cost hours to figure out alone.
- Design consistency — Suggesting hover effects, spacing, and colour usage that fitted the existing style rather than clashing with it.
- Content — Helping shape story drafts into something readable without losing my own voice.
What it isn’t
Claude doesn’t make decisions for you. It doesn’t know what your site should say, or what matters to you, or what you want people to feel when they land on it. That part is still entirely yours.
And there’s a discipline to it: if you don’t understand what it builds, you end up with code you can’t maintain. The goal was always to understand the output, not just ship it.
This site exists because I was willing to build it imperfectly, in the open, with help. That feels like the right way to work in 2026.