Isn't the usual argument that all AIs can do is build on prior art? Like, I spend a disproportionate amount of time trying to convince my agents that I don't want to just reimplement the Rust borrow checker for my language!

- I've built a memory-safe, ownership-based, auto-freeing programming language (https://github.com/andrewjk/nomen) and am starting to use it here and there. It still has a few holes to plug!

- I've also been working on my full-stack JS framework (https://github.com/andrewjk/torpor) which is now faster, has more built-in components, and has single file code routing which can be migrated to full file routing when your site gets bigger

- Both started by a human but now mostly implemented by robots

Multi-language libs with implementations in TypeScript/C#/Swift/Zig so that they can be used directly in web or native apps:

- A Markdown parser/renderer (https://github.com/andrewjk/allmark). The TS version of this was hand written and the other versions are AI translations.

- A Git client (https://github.com/andrewjk/gitologist) and a diffs parser/renderer (https://github.com/andrewjk/differator). These were vibe coded in TS, checked and then AI translated.

I’ve used these libraries to build a native Mac Markdown notes app (https://staunchapps.com/stash) and a JS script launcher (https://staunchapps.com/launch).

Are the creator of the `.mdsv` format that's referenced in the stash manual? Never heard of it before. If you are, interested if there's a spec.

It's just a text file containing a single Markdown table. Making it a different extension seemed like the best way to force them to open in a different editor view inside Stash (while keeping them editable with any other text editor).