> TL;DR: AI assisted coding, when used responsibly, is not banned, however one-shotting something via vibe-coding and hosting it as a complete project, is.
On Tuesday I was working on my Braille 'N Speak emulator[0], an ancient Z180-based accessible notetaker for the blind. I wanted a z180 emulation library that wasn't z180-emu (GPL) and that I could use on the web via wasm, on my phone, and would have a Python binding.
Of course, this is a perfect sort of job for an agent with a verification loop, so I went off and researched if there were any standard conformance suites for z80/z180. Turns out there are a ton! I wired up my verification loops, fired off /goal, and sat back. Now, I have my library[1].
Was "AI assisted coding" "used responsibly?"
Should this sort of thing be allowed? Why or why not?
It's about control. It's always been about control. The moderators of this project would happily throw you under the bus if you ever in any way displeased them while being an AI user.
I think this is the main drive behind this. Make it easier to get rid of people you don't like/don't welcome on your platform without making it explicit, because that could surface some biases of theirs.
Nope, personally this enters full vibe-coding territory for me, and I won't approach this code with a 30 feet pole isolated in a container running in a VM.
Sorry, but seriously.
If it works (and we constantly see LLMs cheat to pass tests), and helps someone, more power to you and them, but no, this is what vibe-coding squarely is, and what Codeberg explicitly excludes.
I see two big problems with that codebase:
- No human, incl. you can modify this meaningfully.
- The AI wrote it won't be able to touch it again, too.
So it's write-once software, and this is the biggest problem with vibe-coded software.
On Tuesday I was working on my Braille 'N Speak emulator[0], an ancient Z180-based accessible notetaker for the blind. I wanted a z180 emulation library that wasn't z180-emu (GPL) and that I could use on the web via wasm, on my phone, and would have a Python binding.
Of course, this is a perfect sort of job for an agent with a verification loop, so I went off and researched if there were any standard conformance suites for z80/z180. Turns out there are a ton! I wired up my verification loops, fired off /goal, and sat back. Now, I have my library[1].
Was "AI assisted coding" "used responsibly?"
Should this sort of thing be allowed? Why or why not?
[0]: https://github.com/ctoth/qns
[1]: https://github.com/ctoth/z-core