I recently built IsMyAIDown.com to challenge some assumptions about modern web development. Instead of reaching for React like everyone else seems to be doing these days, I wanted to see if I could build a distributed, scalable, real-time, single-page app using server-side tech with Phoenix and LiveView.

The core idea was simple enough - monitor major AI providers and show their status across different geographic regions. But the primary goal was to better understand what was possible without the complexity of modern frontend frameworks.

I ended up deploying across three Fly.io regions and using Oban (a very good job processing library in the Elixir ecosystem) to handle geographically distributed health checks. Propogating new checks and detecting and tracking user presence was all quite straightforward. Phoenix is a great framework.

That said, the mental model with LiveView is... a lot. You've got to keep track of a variety of message callbacks that all seem to run together in my head (mount, handle_info, handle_update, handle_params, etc) as well as a heirarchy of views and components that all interact with each other. Even for a relatively simple site like this, it's a lot to hold in your head at once and I found myself having to re-study the flow a few times to get my head around what was going on (surely a personal failing, but a thing none the less). I would not say that LiveView is for beginners.

While I haven't optimized the site much at all, I'd love to see how even this basic implementation holds up under increased load, so do your best HN.

Some side-thoughts on modern software dev: Using Cursor definitely made me more productive but I found myself becoming weirdly disconnected from parts of the codebase that I hadn't written myself. There's something about that line-by-line writing process that builds up your mental map in a way that AI-generated code doesn't quite match. As the size of a system grows, I wonder if AI-generated code will help or hinder?

Quick note on the various AI providers I'm monitoring – OpenAI and Google seem to be the fastest and most reliably available while Anthropic is a bit slower and much more prone to API call failures.

Some resources that I found quite helpful:

Learn Phoenix LiveView: https://arrowsmithlabs.com/ Oban: https://oban.pro World Page Speed Test – planet-wide elastic scale with FLAME: https://fly.io/phoenix-files/world-page-speed-test-elastic-s...