So now that I have a subscription to Claude Code and have poked at it a bit, I felt it was time for a fully vibe-coded project. This is in advance of the librarians’ vibe workshop, which starts on Monday.

I considered this for a few days and ultimately settled on building a Mastodon bot. Other projects were also possibilities: I could revive the Open Journal Matcher, or find a new API to code against, or some such. But I have already made a lot of Mastodon bots (maybe 5-10 over the past few years), so I am familiar with the various steps and effective strategies. I felt I’d be able to judge whether it was doing a good job, and recognize if there were serious red flags. So I opted to build a bot that posts a picture of a saint on their respective saint’s day.

How did it go? Claude initially had trouble with the Wikimedia API, but it was iteratively able to figure it out on its own, and ultimately returned useful data. Watching it work iteratively at this task was in fact quite interesting and remarkable, and is probably the point of this type of tool. Once it had data, it had no problem whatsoever building the bot script. I suppose those things are pretty formulaic and easy to replicate. I asked it for sample systemd configuration files to schedule the bot, but these were less good, probably because I didn’t give Claude the access it needed in this context to iterate independently.

Where is all of this going? I have no idea. But these are interesting times.