I rebuilt my site so an agent can update it

My old site was something I built a while back, and it wasn't very dynamic. I had a lot I wanted to put on it, but nothing was set up so I could share a thought and have an agent put it up. So I rebuilt it with Claude Code, mostly to fix that.

What I want looks like this. I'm scrolling X and see something in product or AI that I have a lot of thoughts on, or questions about. Instead of just retweeting it, I send it to an agent from my phone and it ends up here. A paper trail of what I thought in the moment, that I can come back to. Same goes for a new idea or a side project. Maybe the agent even builds a rough proof of concept.

The first version was pretty lame.

First pass. Generic, like most first passes.

That's normal. When you start something new with AI and share your ideas without much context, you get something generic back. You fix it by giving it more: context, examples, things you like from around the web. That's what got it to a starting point I felt good about.

A lot of the work after that was cutting. The model kept adding stuff, especially when it worked from my resume, and most of it was slop.

Where it landed. Simpler on purpose.

The part I care about is behind the page. The repo has instructions the agent reads first: how I sound, how a note is put together, how to publish without breaking anything.

What the agent reads when I send it something.

I haven't had the moment where it just gets it yet. It's new, and I'm sure a lot of this changes in a few weeks. Honestly the friction isn't the tool. It's me being impatient when I know exactly what I want and can't quite say it.

I'm also not handing it the keys. For that it would have to read my mind and know what I find interesting out there, and I don't think we're close.

If you're thinking about setting up something like this, just try it. Failing is the fastest way to figure out what works.