Back in October, Paul Ford wrote this brilliant piece that I think 100% sums up my feelings about the current state of AI and tech more broadly speaking. The hype may be at insuferable levels right now, but what is being unlocked for the nerds and script kiddies who build things for fun is kinda nuts, and this is the kind of stuff that will continue to thrive when the bubble bursts.
This weekend I vibe-coded a little project that's been rattling around in my head for a while, but given my limited skill level would have in the past just reamined an idea, or taken months and months of work: an API for tracking art exhibitions in New York City.
If you're an art nerd who has ever lived in NYC, you know that there are so many art exhibitions happening at any given moment. Keeping track of what's open, what's closing soon, and what you can't is a bit impossible.
Rather than build myself a one-off tool, I figured why not build the information gathering engine and put a free public API in front of it. That way, not only can anyone use it, but I can integrate the info into whatever I want, some ideas:
- An e-ink screen on my fridge showing what's closing this week
- Automated alerts when specific artists have shows
- Integration into personal websites or apps
- Tools other people build that I haven't even thought of
Plus, it's free. Because if I'm going to solve this problem for myself, I might as well solve it for everyone.
Vibe Coding It Out
I banged out the whole thing in a weekend with the Claude Code CLI, working on it just here and there, getting chores done while Claude worked on building, testing, and deploying.
Some venues have pristine structured data. Others have exhibition info buried in paragraph text. Some list "ongoing" with no end date. Some don't list dates at all. So rather than deal with parsing and normalizing all of the messy data in some kind of labor intensive hard-coded manner it offloads that to the Anthropic API with a two-pass process, first trimming the raw HTML down to just a list of shows and their info, and second structuring all of that in the JSON format we need.
It's wild to me how advanced coding models are getting — as a nerd who always has little ideas for utilites I could build to automate certain things, or just make fun tools to make my life a little easier, I can now just spend a bit of time laying out the specs, and then whip it up in a weekend. And while I mentioned Claude Code, the offline and open source models and tools are also insanely good now.
It's funny, in the 2010s everyone was saying "learn to code", and now, it's something more like "learn how to conceive of and architect modular services" because the wrote tediuous parts can be handled for you — you just need to project manage it a bit.
What's Next?
The crawler is functional but definitely needs work. Right now it hits the major museums and galleries, but the logic for extracting and normalizing exhibition data could be a lot smarter. There's also the question of update frequency,and caching strategies. For now it just crawls weekly, which honestly may be enough. Some more to-do items:
- Smarter content extraction and validation
- More venues (there are hundreds in NYC)
The docs are here and please do let me know if you build something with it. I'd love to see what people come up with.
I also built a simple demo client to test the API in action:
