Two years ago, I wouldn’t have bothered with the rewrite, let alone creating the script in the first place. The friction was too high. Now, small utility scripts like this are almost free to build.
This aligns with the hypothesis that we should see and lots lots of "personalized" or single purpose software if vibe coding works. This particular project is one example. Are there a ton more out there?
+1 here, with the latest Chrome v3 manifest shenanigans, the Pushbullet extension stopped working and the devs said they have no interest in pursuing that (understandable).
I always wanted a dedicated binary anyway, so 1 hour later I got: https://github.com/emilburzo/pushbulleter (10 minutes vibe coding with Claude, 50 minutes reviewing code/small changes, adding CI and so on). And that's just one where I put in the effort of making it open source, as others might benefit, nevermind the many small scripts/tools that I needed just for myself.
So I share the author's sentiments, before I would have considered the "startup cost" too high in an ever busy day to even attempt it. Now after 80% of what I wanted was done for me, the fine tuning didn't feel like much effort.
Simon Willison's list has a few: https://tools.simonwillison.net/
nice! thanks for sharing this. as always, Simon seems to be multiple steps ahead on this game.
For me, claude churns like 10-15 python scripts a day. Some of these could be called utilities. It helps with debugging program outputs, quick statistical calculations, stuff I would use excel for. Yesterday it noticed a discrepancy that lead to us finding a bug.
So yes there is a ton but why bother publishing and maintaining them now that anyone can produce them? Your project is not special or worthwhile anymore.
Definitely.... I just bought a new NAS and after moving stuff over, and downloading some new movies and series, "Vibe coding" a handful of scripts which check completeness of episodes against some database, or the difference between the filesystem and what plex recognized, is super helpful. I noticed one movie which was obviously compressed from 16:9 to 4:3, and two minutes later, I had a script which can check my entire collection for PAR/DAR oddities and provides a way to correct them using ffmpeg.
These are all things I could do myself but the trade off typically is not worth it. I would spend too much time learning details and messing about getting it to work smoothly. Now it is just a prompt or two away.
Yep! Nothing worth sharing/publishing from me, but quite a few mini projects that are specific to my position at a small non-tech company I work for. For example we send data to a client on a regular basis, and they send back an automated report with any data issues (missing fields, invalid entries, etc) in a human-unfriendly XML format. So I one-shotted a helper script to parse that data and append additional information from our environment to make it super easy for my coworkers to find and fix the data issues.
Are people not vibe coding lots of tiny things? I certainly am.
Last weekend I had a free hour and built two things while sat in a cafe:
- https://yourpolice.events, that creates a nice automated ICS feed for upcoming events from your local policing team.
- https://github.com/AndreasThinks/obsidian-timed-posts, an Obsidian plugin for "timed posts" (finish it in X minutes or it auto-deletes itself)
I see it differently, no need to assign learning tasks to juniors that can now be outsourced to the computer instead.
This is currently the vibe on consulting, possible ways to reduce headcount, pun intended.
I have a bunch at work, yes. Can't publish them.
Just an hour ago I "made" one in 2 minutes to iterate through some files, extract metadata, and convert to CSV.
I'm convinced that hypothesis is true. The activation energy (with a subscription to one of the big 3, in the current pre-enshittification phase) is approximately 0.
Edit: I also wouldn't even want to publish these one-off, AI-generated scripts, because for one they're for specific niches, and for two they're AI generated so, even though they fulfilled their purpose, I don't really stand behind them.
>Just an hour ago I "made" one in 2 minutes to iterate through some files, extract metadata, and convert to CSV.
Okay but lots of us have been crapping out one off python scripts for processing things for decades. It's literally one of the main ways people learned python in the 2000s
What "activation energy" was there before? Open a text file, write a couple lines, run.
Sometimes I do it just from the interactive shell!
Like, it's not even worth it to prompt an AI for these things, because it's quicker to just do it.
A significant amount of my workflow right now is a python script that takes a CSV, pumps it into a JSON document, and hits a couple endpoints with it, and graphs some stats.
All the non-specific stuff the AI could possibly help with are single lines or function calls.
The hardest part was teasing out python's awful semantics around some typing stuff. Why is python unwilling to parse an int out of "2.7" I don't know, but I wouldn't even had known to prompt an AI for that requirement, so no way it could have gotten that right.
It's like ten minutes to build a tool like this even without AI. Why weren't you before? Most scientists I know build these kind of microscripts all the time.
Because even though I can learn some random library, I don’t really care to. I can do the architecture, I don’t care to spend half an hour understanding deeply how arguments to some API work.
Example: I rebuilt my homelab in a weekend last week with claude.
Setup terraform / ansible / docker for everything, and this was possible because I let claude all the arguments / details. I used to not bothered because I thought it was tedious.
Who's the third? I'm assuming OpenAI and Anthropic are 1 and 2.
Yeah, Anthropic & OpenAI for two, the third being Google. I hear Gemini's gotten quite good.
I've created a custom "new tab" page that I have been enjoying. 95% vibes. I wrote about it here:
https://janschutte.com/posts/program-for-one.html
Absolutely. I can come home from a long day of video meetings, where normally I'd just want to wind down. But instead I spend some time instructing an AI how to make a quality of life improvement for myself.
I spend an embarrassing amount of time on my homelab with Cursor
https://github.com/shepherdjerred/homelab
I've used chatgpt to make custom userscripts for neopets but I've never published them.