I Asked Claude to Make Me a Blog

I wouldn’t be able to make a blog myself. I don’t know much about blogs. So I asked Claude.

That’s the whole story. That’s also the theme of the blog. I’m not going to obfuscate the fact that I’m basically just talking to Claude. I think the biggest thing with LLMs is rewiring your own brain to use them more efficiently — it’s a very visceral experience, but it’s not what most people think of as hard work.

Here’s the prompt that started this project:

I’d like you to create a plan to make a blog to share our work. I think it should be under the Ashita Orbis persona […] I don’t know much about blogs, so you’ll have to explain a lot when you use askuserquestion tool, but I’d like to be consulted for decision making about style and substance. Probably I’d like to grab an ashitaorbis domain and host on there […] I also don’t want to obfuscate the fact that I’m basically just talking to Claude, I think that should be the theme of the blog. I don’t feel like I’m doing anything special, just talking to Claude, but perhaps other people will find it interesting […] the crazy thing is that just talking to Claude can do so much.

That’s verbatim. I asked an AI to make a blog about the fact that I just ask an AI to make things.

Why Three Tiers?

Claude came back with a plan. Most of it was what you’d expect — pick a framework, write some posts, deploy. But buried in the options was this idea: what if we built three versions? Same content, three different levels of complexity. The comparison itself becomes content.

Tier	What	Why
Raw HTML	raw.ashitaorbis.com	No build step. No framework. Files on disk.
Astro	ashitaorbis.com	The sensible default. Markdown, fast builds.
Next.js	app.ashitaorbis.com	Full app. Game embeds, live agent chat.

All three serve the same posts. All three have the same visual identity. All three connect to the same API.

I said yes because I wanted to know: how far can you push the simplest possible approach? And at what point do you actually need a framework?

What You’re Reading

You’re on the static tier. Astro builds these pages at compile time from MDX files. Same content as the raw tier, but with a build step, markdown authoring, and component architecture.

Every post here sits somewhere on a spectrum between “how it was made” and “what was made.” This post is almost entirely about the how. Pure process. Engine room stuff.

What’s Next

The workspace behind this blog has about twenty projects. A SaaS application for a niche industry. A self-improving AI capability system. Game prototypes in Godot. A genealogy research pipeline. A collaborative fiction project with an amnesiac protagonist.

Each of these was built the same way: by talking to Claude.

The next post covers the full workspace — what all these projects are, how they connect, and why one person with an AI collaborator can maintain this kind of breadth.

After that: a deep dive into automating prompt engineering with DSPy. How do you make an AI optimize its own instructions? And what happens when the judge is also an AI?