Rebuilding a personal site is one of those things developers circle back to every few years. What started as a “quick redesign” spiraled into a full stack rebuild, new frameworks, new hosting, and a bunch of AI experiments I didn’t see coming. Turns out, giving GPT a seat at the design table makes for a faster (and way more interesting) build.

The stack

But before we get into the AI side of things, let’s talk about the tools holding it all together, the stack that’s powering my new website.

• Next.js App Router

  I have worked with the Pages Router before. This time, I’ve been exploring the App Router with React Server Components (RSC). It’s been solid so far. Nested layouts make it easier to structure pages, and routing feels more flexible, especially with route groups. The built-in caching and streaming defaults have saved me time, and overall, it’s handling the kind of site I’m building without much hassle.

• Tailwind CSS

  I have used vanilla CSS, Sass, and Material UI before. Tailwind has been the easiest to learn and integrate into my workflow. Since I had already used it at Thera, I’ve been comfortable with the approach. Utility-first styling is keeping things fast. I’m styling directly in markup without switching files or managing class naming.

• MDX

  Markdown has been great for writing. MDX is making it flexible. I can drop in custom components where needed, without leaving the writing flow. That’s how this post is working, just content and JSX, living together.

• AWS Amplify

  I have used the traditional S3 + CloudFront setup before. It worked, but I ended up spending a lot of time setting up buckets, policies, CI/CD, and invalidations.

  This time, I’ve been using AWS Amplify. I haven’t had to think about infrastructure. I’ve just been pushing code.

  It hasn’t been as configurable, but for personal projects, that’s a tradeoff I’m fine with.

  That said, I did hit a blocker while trying to set up a separate Amplify project for the blog at blogs.sahilten.com, and then rewrite it to serve under sahilten.com/blogs. The goal was to keep the main site public, but keep the blog project private so if I added features like a like counter, it wouldn’t be as easy to target with scripts or automated abuse.

How AI fits into the workflow

• Design help from Galileo AI

  After trying a few different tools to help with design Galileo AI stood out the most to me.

  Its output wasn’t perfect, but it gave me a solid base. The designs were clean, responsive, and built on common UI patterns I could extend.

  I didn’t ship them as-is. I tweaked spacing, restructured components, and added my own style. But having a strong starting point made a big difference. I spent less time guessing and more time building.

• Development with Cursor

  I’ve been using Cursor as my main editor. It’s built on VS Code but adds a fast, context-aware AI layer.

  The Tab to autocomplete feature has been a standout. It’s saving time on props, boilerplate, and repetitive logic. I’m accepting what works, adjusting what doesn’t, and moving on.

  I’ve also been using it for refactoring and quick explanations. It’s not thinking for me but it’s keeping me moving.

My Takeaways

The tools are better. The support is smarter. If you’re curious and willing to build, there’s never been a better time.

AI has made it easier than ever to learn by doing, especially when I’m figuring things out on my own.

I’m asking questions, getting targeted answers, and unblocking myself without falling into a doc rabbit hole. It’s lowering the friction between idea and execution.

However, I did notice that it’s almost too easy to rely on it too much. I started letting it write functions, fix bugs, even draft comments. It saved time, but it also cost me depth.

When I’m struggling through a problem, I remember it better. I’m building intuition. That’s how real learning sticks.

AI is a great assistant. But it’s not for replacing the thinking part. That’s still where I’m growing.