Today I sat down with Claude and rebuilt my entire portfolio site from the ground up. What started as a simple Express.js scaffold turned into a VS Code-inspired interface, complete with playable Snake and Chess games. The whole thing took a single day.
I want to talk about what that experience was like, because I think it reveals something interesting about where AI-assisted development is headed.
The Process Was Iterative, Not Linear
I didn't start the day knowing I wanted a VS Code theme. I started with a minimalist black-and-white aesthetic—Courier font, sharp edges, stripped down to essentials. That was fine, but it felt generic. So I kept pushing.
What's interesting is how the iteration happened. I'd describe a vague direction, Claude would generate something concrete, and then I'd react to what I saw. My taste was the filter; the AI was the engine. We went through multiple passes—adjusting the favicon, tweaking font weights from 700 down to 100, adding a sidebar file tree, then eventually landing on the full IDE aesthetic.
This isn't how I usually work. Normally I'd sketch something in Figma first, then implement it. But with Claude, the implementation was the sketch. That changes the feedback loop entirely.
Adding the Games
Somewhere in the middle of the day, I realized the site felt too static. It was a portfolio, sure, but it didn't have any personality. So I asked for Snake. Then Chess.
Both games work. Snake has the classic mechanics—arrow keys, growing tail, game over on collision. Chess has turn-based play with move validation and a basic AI opponent. They're styled to match the IDE theme, with the file tabs reading snake.js and chess.js.
Could I have written these myself? Sure, eventually. But getting two fully functional games integrated into my site's design system in under an hour? That's a different kind of leverage.
What I Actually Learned
A few observations from the day:
Direction matters more than specification. I didn't write detailed requirements. I pointed at a direction and let Claude fill in the gaps. When the output wasn't right, I course-corrected. This felt more like creative collaboration than traditional development.
Reading code is still essential. Everything Claude generated, I read. I caught edge cases, made stylistic adjustments, and occasionally rewrote sections entirely. The AI doesn't replace understanding—it accelerates the parts that don't require deep thought.
Personality comes from iteration. The site has character now because we went through multiple versions. The first pass was boring. The fifth pass had a point of view. You can't shortcut taste, but you can compress the time between attempts.
Where This Goes
I wrote a post a while back about using AI tools without losing your edge. Today reinforced that perspective. I'm not worried about becoming dependent on these tools because I'm still the one making decisions, still the one with opinions about what's good. The AI just helps me get there faster.
But I'll admit—building an entire site with games in a single day does make me wonder what's possible when you really lean into this workflow. I'm curious to find out.