I’ve been pretty sick the last couple weeks, and during some top-tier doomscrolling I stumbled onto a little browser game called 100 Jumps. It’s one of those “why is this so hard” games where you hold space to charge, release to jump, and try to make it across platforms without falling. After wasting way too much time on it, I decided to see what would happen if I asked AI to build my own version from scratch—so I opened Claude, dropped a single prompt with the link, and let it rip. What I got back was honestly wild: a full HTML canvas game with a title screen, physics, trajectory preview, particle effects… the whole deal. From there, I basically just described what I wanted—moving platforms, sound effects, a perfect-landing streak that earns extra lives, and eventually enemies like orbiters and chasers—and I only stepped in to report bugs. Once it felt good, I had Claude package it into Docker (NGINX, compose, entrypoint, even an env var for the site title), pushed it to my self-hosted Git, and deployed it on my server with a domain. And that’s where my concern kicks in: this took about an hour, I wrote zero lines of code, and that’s a problem for self-hosting. If non-devs start shipping AI-built apps with no real code review or security mindset, we’re going to get flooded with projects that are risky, unsupported, and dead-on-arrival the second something breaks.