logoalt Hacker News

qwoprunner • today at 5:59 PM • 1 reply • view on HN

The lack of much-improved open source tools is one of the more baffling things for me when it comes to AI. Especially after Sol 6.1 / Opus 5.5 released, you can whip up a fairly solid application in several days. But... where are they?

OSS devs were sinking a mountain of time into relatively unpaid work, doing unglamorous work. Some of it is in backend systems and console utilities that few people will ever appreciate, but a lot of effort went into user-facing things. And yet, so far I found just one program that replaced one of my mainstays - I went with AI-coded SquirrelDisk to replace FileLight. Everything else is the same, and I just don't understand that. Surely not EVERYBODY rolls their own? Surely some OSS people did embrace AI rather than quitting in disgust?

A small example: tagging music files on Linux is horrible. Kid3 is.. it works, but the UX is almost like a parody. Mp3tag runs on windows and it has a macos version and it's a vast improvement. You can run it through bottles but I refuse to believe everybody does that. Surely people that use Linux listen to music, which leads itself to occasionally needing to tag or rename your audio in batches.

I had some spare time so I rolled my own mp3tag clone that would work on Linux, MacOS and Winblows. I added audio playback to it because opening in external program annoyed me. Then I added some functionality for podcasts to it: being able to search for subtitles, playback speed and such. Then I added the ability to convert between formats, using ffmpeg. Then I added a way to measure loudness, some ReplayGain functionality and the like. Then a way to find duplicates, for making sure those random files you have on that old external HDD can be deleted safely, and a way to search through what you have on all disks, even if they're disconnected, so you don't buy some album you already own like a dumbass.

It took me about three weekends of tinkering with it until it was good enough to switch to it. I keep running into small issues but by and large it's been fairly reliable. It's fairly complex, with a lot of buttons in the view, but given that the 'competition' is mp3tag (which is okay but also ugly) or kid3 (which is both ugly and almost unusable) it's just flat out superior, unless you're happy with mp3tag and don't want all the extra complexity. I don't want to offer unpaid tech support to people on GitHub but I'm considering releasing it and just leaving it there with no further communication and maybe somebody will pick it up. If not, it's still better than alternatives even if there are undoubtedly many, many bugs and the codebase is undoubtedly a tangled mess.

Aside from the ugliness of having almost all functionality without any sub-menus that would give Apple designers heart attacks and perhaps more complexity than it should have (I personalized it for what I wanted, after all), it's genuinely a decent application. It's written in c++, a language I never used. It's fairly performant on my ~15k audio files library. Changing it is as simple as pointing a modern LLM at it and asking in english for some changes. It took me approximately 20 hours to make and the cost was about $10 in subsidized tokens from Claude.

If I, a random guy who doesn't even know c++, can write a usable program in three weekends, then where is all the new vibecoded software? There are so many tools that technically work but the user experience is straight up awful, usually on linux side, or flat out holes that aren't filled, usually on mac side. As best I can tell, very little is being developed. People who have no idea about coding are recompiling console games on pc, in admittedly dubious quality, but people that dedicated years of their lives to OSS... don't?


Replies

Kim_Bruning • today at 6:27 PM

I'm running into this with just my online friends already. We each have wildly diverse ways of coding now, and our programs don't seem to want to interact with each other. It's like the lisp curse. It's so easy to make your own custom code that you don't bother to even look at what the neighbors are doing.

As AI build-systems become more complex, it might be interesting to publish their work in central locations, and build new standards by which these kinds of software can interact. Or adopt existing ones in new ways!

Right now it's not always clear when and where you can apply an AI workflow and under which rules. This needs to be coordinated sooner or later.