I built my portfolio site twice.

The first time I asked AI where to start. It suggested Astro. Lightweight, modern, perfect for small static sites. Made sense to me. I went with it, vibe coded my way through it, stopped long enough to make sure I actually understood what I was looking at. And I ended up with something I was genuinely happy with. Then I scrapped the whole thing. I rebuilt the exact page in vanilla HTML and CSS. Page loads dropped from ~150ms to ~22ms. Seven times faster. I'll be upfront though, nobody was suffering at 150ms. The more interesting number is the 80% reduction in code. One file. No build pipeline. No dependencies. Just a page. Here's the thing... AI recommends what's statistically common. Frameworks are the right answer often enough that suggesting one is reasonable. We know this. But when we're not comfortable enough to ask questions we trust the process. I was building a single page that looks almost identical to my resume. I didn't need a framework. I needed an afternoon and possibly a caffeinated beverage. AI is an incredibly versatile hammer. The trick is remembering not everything is a nail. There's a version of using these tools where you stop asking yourself whether the suggestion actually fits. That version will quietly make your life more complicated over time, at any scale. How many nines did AWS have last month? Thankfully mine was a portfolio website nobody visits. Not a multi-billion dollar service millions of people rely on. The Astro version wasn't wasted time either, for what it's worth. Understanding why something is overengineered is its own kind of useful.