Starting over on my website, and why the first one died
I let an AI write my entire site. It shipped in a weekend and I never opened it again.
Two months ago I had a personal website. It existed, it loaded, the layout held up on a phone. By any reasonable definition I had shipped something. I described almost none of it to anyone, added nothing to it, and last week I could not have told you how its navigation was built without opening the files and reading them.
So I deleted it and started over. This post is me trying to say why, in a way that is useful rather than just self-critical.
The obvious diagnosis is wrong
The easy story is "AI wrote bad code." It did not. The code was fine. Cleaner than what I would have produced on my own, probably.
The problem was that I could not find my way back into it. Every file was a set of decisions somebody else had made for reasons I had never been asked to consider. When I opened the folder to add a project, the first task was not "write the project." It was "reverse-engineer my own website." That is a much larger task, and it is the kind of task you keep deciding to do tomorrow.
A site you do not want to open is not a site. It is a screenshot with a domain name attached.
What I think actually happened
I optimised for the wrong finish line. I treated "live on the internet" as the goal, and by that measure v1 was a total success. But the actual goal was a place I would keep adding to for years. Against that measure it failed on day three and I did not notice for two months.
The other thing worth naming: generating the site did not remove the work of understanding it. It moved that work to a later date, one I had not scheduled and did not want to spend a Saturday on. I felt fast in the moment and paid for it in a currency I did not realise I was borrowing.
What I am doing differently
Same tool, different job. I am still using AI constantly on v2, but the request has changed from "build this" to "explain this, then watch me build it." When I ask why a page uses <article> instead of <div>, I get an answer I keep. When I ask for a finished page, I get a page I own on paper and not in my head.
Three rules I am holding myself to:
- No file I cannot explain. If I could not walk a friend through what a line does, it does not go in.
- Smallest version that works, then add. v1 had features I never once used. v2 launches with three pages.
- Every addition goes through the full workflow. Branch, commit, merge. Even for a typo. The point is the reps, not the ceremony.
The uncomfortable part
I do not know yet whether this works. That is the honest position. It is entirely possible I am rationalising a rebuild I wanted for other reasons, and that in three months this site is as dormant as the last one.
The difference is that this time the record is public and dated, so if it goes quiet, that is visible too. Which is roughly the point of building in public: it makes the failure mode legible instead of private.
Check back and see.