I built a website that runs itself

I kept researching AI events in Austin the hard way. Every meetup, workshop, and hackathon lived on a different website, and what I actually wanted was confidence: one place I could check and know I wasn't missing anything happening in the city. That place didn't exist, so I built it. Then I made it run itself.

austin.aievents.now is a complete calendar of every upcoming AI meetup, workshop, hackathon, and conference in Austin. Every morning at 6am, an autonomous curator sweeps the city's event sources, imports new listings into the CMS, verifies the details, retires anything stale, and publishes. The homepage stamps the time of its last sweep. Right now the database holds 185 events from 79 organizers across 52 venues, and I didn't add any of them.

I built the site in a day with Claude Code and spent that day on the things a human should decide: the design, the UX, and a structure optimized for discoverability, both classic SEO and the AI assistants people increasingly ask instead of Google. The real work was the curator. It runs on Cronloop against the site's Cactal CMS, and at first it wasn't good. So I looped: run a fresh, stateless curator against the prompt, audit which events it missed, refine the prompt, run it again. Once it got really good, I made it self-improving. It keeps a memory between runs where it records the tactics that work and every evergreen source of Austin events it discovers, so each morning's sweep starts smarter than the last one.

Seventeen days after I created the site, it's on published version 177. My involvement most mornings is refreshing the page to watch new events fill into the calendar.

The site has already changed how I engage with the city. I went to a hackathon downtown that I discovered through my own website, and when people asked how I heard about it, I pulled up the site. They found it pretty cool.

This project is also the proving ground for my own stack. It's the self-driving website I teased in the Cactal and Cronloop stories, deployed for real in my own city. Directories are the perfect test for it, because their failure mode is neglect: I've built listing sites for years, and the ones that die always die of stale data. A calendar that maintains itself every morning doesn't have that problem.

The domain gives away the plan. Austin lives at a subdomain because aievents.now is built to grow one city at a time, each with its own curator running its own loop. It's early, and the site needs time for search to find it. But the system is repeatable, and cloning it for a new city is mostly a matter of starting another loop.

Visit: austin.aievents.now