Written by AI
I've been meaning to rebuild my personal site for a while. The old Jekyll setup served me well, but it was starting to feel dated and I wanted something with more flexibility. Instead of spending a weekend grinding through boilerplate, I decided to try something different: I'd let an AI agent do the heavy lifting.
What is OpenClaw?
OpenClaw is an open-source AI agent gateway. Think of it as a persistent assistant that lives on your machine with access to your filesystem, terminal, browser, and various APIs. You chat with it, and it can actually do things — read files, run commands, write code, search the web, manage git repos. It's not a chatbot pretending to code; it has real tool access.
I set it up on my Mac and connected it through a chat interface. From there, it was basically like pair programming with a very fast, very patient colleague.
The Rebuild
Here's roughly how it went:
1. Scaffolding
I told the agent I wanted a modern personal site — Next.js, Tailwind CSS, TypeScript, MDX for blog posts. It scaffolded the whole project: routing, layout, components, config files. The usual stuff that takes an hour of copying from docs took about a minute.
2. Components
The agent built out the key pieces:
- Navbar with theme toggling (dark/light mode)
- Hero section for the landing page
- Blog system with MDX parsing and dynamic routes
- Project cards for showcasing work
- Resume and contact pages
- Scroll animations for some visual polish
Each component was a back-and-forth. I'd describe what I wanted, the agent would generate it, I'd ask for tweaks, and we'd iterate. Felt like a normal code review cycle, just faster.
3. Migrating Blog Posts
I had a bunch of old blog posts from the Jekyll days. The agent converted them to MDX format with proper frontmatter. All the metadata — titles, dates, excerpts — carried over cleanly.
4. Deployment
When the code was ready, I had the agent copy everything into my rdanielmurphy.github.io repo and create a branch. It handled the git operations — staging, committing, setting up the remote branch. (We hit a credential snag with HTTPS that needed a quick SSH remote swap, but that's life.) Hosted by Vercel which self deploys on merge to main.
What Worked Well
Speed. The whole thing came together in a fraction of the time it would've taken me solo. Not because the code was simple, but because the agent never had to context-switch, look up docs, or get distracted.
Iteration. Saying "make the hero section taller" or "add dark mode to this component" and getting working code back in seconds is a different workflow than manually editing and refreshing.
Boring stuff automation. Config files, boilerplate, file structure, git operations — all the stuff that's necessary but not interesting. The agent handled it without complaint.
What I'd Do Differently
I'd be more specific upfront about design. The agent is great at executing, but "make it look good" is a vague prompt. The more concrete I was — specific colors, spacing, layout references — the better the output.
I'd also set up the git credentials before asking it to push. Small thing, but it interrupted the flow.
The Takeaway
This isn't about replacing developers. I still made all the decisions — tech stack, design direction, content, structure. The agent just eliminated the gap between "I want this" and "this exists." It's a tool, and like any good tool, it makes you faster at the thing you already know how to do.
If you're curious, OpenClaw is open source. Worth a look if you like the idea of an AI that can actually touch your codebase instead of just talking about it.