3 ms·

Isn’t that a door (haha) opened by our supreme new AI coding overlords: that now it’s possible to take the Every Door source code (it’s on Github), and throw it

Isn’t that a door (haha) opened by our supreme new AI coding overlords: that now it’s possible to take the Every Door source code (it’s on Github), and throw it into a coding agent with the instruction to „improve <ux shortcoming of your choice> without breaking functionality“, and with some iterations and human validation that yes, the UX is now indeed better, play the changes back with a PR?

raybb 14d ago
I can improve the UI to my taste but if the maintainer doesn't have time/money to maintain a big ol vibe coded PR for some very opinionated change isn't going to do much. That would appear to leave me with the choice of just using it for myself (lame) or I then have a public fork to maintain :( If there's basically no shot of it getting up streamed I think I'd rather make something from scratch to suit my needs without the extras. Anyway, I hope the AI overlords help us fund more maintainers soon!
- senbrow 13d ago
  FYI LLMs are also very good at maintaining a fork ;)
  - raybb 13d ago
    Do you have any examples of forks maintained by LLMs and how they do it?
    - DeepDyno 13d ago
      I keep my neovim config maintained by a LLM which is a fork of Kickstart. My example might be an outlier on the easy side because they have a custom/plugins directory that will not have merge conflicts ever so I try to never modify the main init.lua and keep all of my changes separate in my own files. This prevents merge conflicts. I can then ask my LLM to add a plugin or customize existing plugins by writing files in this separate directory to add/overwrite plugins and configurations. This works great for me. If I were working in the main file creating merge conflicts, I could imagine it could get messy fast.
Gigachad 13d ago
UX is one of the things AI isn’t good at. You can ask it for specific changes but just “make it nicer” is unlikely to work.
- katbyte 13d ago
  No but if you have specific asks like “make this resizable” “I should be able to drag and drop to combine” or “add a button on the top right to close” it’s quite good at directed changes
- mmahemoff 13d ago
  “Make it nicer” will lead to unpredictable and controversial changes, likely to clash with the project owner’s vision. The owner would be better off using the LLM as a collaborator, if anything, to decide where to take the design. However, UX can also be improved by some simple heuristics that could be seen as mechanistic by the standards of modern LLMs. Performance and responsiveness, for example, contribute to UX even when the UI remains fixed. I’d wager Fable-level LLMs could improve perf on most open-source GUIs by benchmarking typical use cases and progressively applying standard tricks like caching, buffering, batching network calls.
  - Gigachad 13d ago
    That's real work and requires real skill. The parent comment example prompt just doesn't work.
- tyre 13d ago
  Claude Design can help. Fable is pretty good at UI when you narrow down to a specific page or—better—something like a modal, table, etc. If someone is a user, they'll have specific complaints. Claude Design is more than capable, which can then be fed directly into Claude Code to implement, test, iterate, and ship. I agree that it lags behind coding and imo the highest leverage area for Anthropic to invest in. Hopefully they do so; UI is consistently one of the main (severe) differentiators between OSS and paid options.

raybb 14d ago

I can improve the UI to my taste but if the maintainer doesn't have time/money to maintain a big ol vibe coded PR for some very opinionated change isn't going to do much. That would appear to leave me with the choice of just using it for myself (lame) or I then have a public fork to maintain :( If there's basically no shot of it getting up streamed I think I'd rather make something from scratch to suit my needs without the extras. Anyway, I hope the AI overlords help us fund more maintainers soon!

senbrow 13d ago
FYI LLMs are also very good at maintaining a fork ;)
- raybb 13d ago
  Do you have any examples of forks maintained by LLMs and how they do it?
  - DeepDyno 13d ago
    I keep my neovim config maintained by a LLM which is a fork of Kickstart. My example might be an outlier on the easy side because they have a custom/plugins directory that will not have merge conflicts ever so I try to never modify the main init.lua and keep all of my changes separate in my own files. This prevents merge conflicts. I can then ask my LLM to add a plugin or customize existing plugins by writing files in this separate directory to add/overwrite plugins and configurations. This works great for me. If I were working in the main file creating merge conflicts, I could imagine it could get messy fast.

Gigachad 13d ago

UX is one of the things AI isn’t good at. You can ask it for specific changes but just “make it nicer” is unlikely to work.

katbyte 13d ago
No but if you have specific asks like “make this resizable” “I should be able to drag and drop to combine” or “add a button on the top right to close” it’s quite good at directed changes
mmahemoff 13d ago
“Make it nicer” will lead to unpredictable and controversial changes, likely to clash with the project owner’s vision. The owner would be better off using the LLM as a collaborator, if anything, to decide where to take the design. However, UX can also be improved by some simple heuristics that could be seen as mechanistic by the standards of modern LLMs. Performance and responsiveness, for example, contribute to UX even when the UI remains fixed. I’d wager Fable-level LLMs could improve perf on most open-source GUIs by benchmarking typical use cases and progressively applying standard tricks like caching, buffering, batching network calls.
- Gigachad 13d ago
  That's real work and requires real skill. The parent comment example prompt just doesn't work.
tyre 13d ago
Claude Design can help. Fable is pretty good at UI when you narrow down to a specific page or—better—something like a modal, table, etc. If someone is a user, they'll have specific complaints. Claude Design is more than capable, which can then be fed directly into Claude Code to implement, test, iterate, and ship. I agree that it lags behind coding and imo the highest leverage area for Anthropic to invest in. Hopefully they do so; UI is consistently one of the main (severe) differentiators between OSS and paid options.