Vibecoding Personal Utilities
Mar. 4, 2026
Wes McKinney’s blog makes a good case for the go programming language as a complement to LLMs. Like Wes, I do not know any go, but somehow it worked!
Right now, people have plenty of opinions and theories on what works best when coding with agents (test-driven development, tight feedback development loops, or some specific prompting frameworks). I don’t have the engineering experience to predict which of these principles will hold up against the pace of top models. But I do have enough experience to know which problems I want computers to solve for me.
I want to share a snapshot of some bespoke projects that I have achived relatively easily as of March 2026. I am curious about the frontier of what is doable now that wasn’t before, and whether we’re going to see a lot more people will building genuinely useful, DIY software.
Here are three single-use utilities I recently “vibecoded”:
podcast-go : I recently started swimming and I like to listen to podcasts while I do. This requires uploading mp3 files to my bone-conduction headphones (a real throwback). Parsing RSS feeds every couple of days to get the latest mp3 is tedious, and podcast hosts don’t make direct file access easy. I’m sure existing solutions are out there, but that’s the point: I didn’t have to look for one. I wrote an app for exactly my use case in a couple of hours with light supervision. This was my first time using Go, which happens to have fantastic Terminal User Interface (TUI) libraries.
esync : I spend a lot of time writing code locally, but sometimes I need to execute it remotely. I used to keep a bash script to selectively sync files, but eventually, I’d forget to run it, get stuck on a bug, and realize I wasn’t even editing the remote code. Yes, there are easier solutions (like Emacs Tramp), but writing this custom utility (a wrapper around
fsnotifyandrsyncinside a Go TUI) removed the friction from my specific workflow.wrds-dl : This serves as both a quick explorer tool for WRDS databases (via their Postgres system) and a command-line tool to download the data without writing or loading any Python. I also generated a skill to make it easy to tell Claude: “Download the monthly stock file from WRDS between 1980 and 2020 and show me that the skewness is all gone! .” This one was honestly more fun than it was strictly useful!
This will affect research in uncertain ways that might not all be fun. But I want to be positive and optimistic to what it brings us in building simple things.
