Writings

LLM as a Web Server

I built a site where every new page is generated live by an LLM instead of a template. Here's what happens when the model is the entire UI layer.

Anthropic just confirmed two of my key suspicions about why Mythos took so long to ship.

Subs aren't a rug pull waiting to happen, they're how frontier labs soak up spare inference capacity.

Anthropic gating Mythos for safety alone doesn't add up. Here's what I think is actually going on.