Develop a small API that accepts a prompt, starts a sprite sandbox, and then you wait for the AI agent to open a pull request.
I developed that in less than a week and now my client owns the code, can customize it to their needs, and if they want to use a different compute environment, they can with minimal code changes needed.
I'm sure everyone has their own tolerance for what is and isn't maintainable :). For me, not knowing what code exists, where it is, how it fits together, and stuffing it all in one main file feels like a recipe for trouble down the road. Sure, I could probably tell the LLM to split the main file into modules and ask it to refactor code etc.
However, from personal experience I'm a lot more efficient when I use LLMs to help with tedious, boilerplate-like code writing but I remain in control over structuring the project so it's maintainable by more than machines only.
I use LLMs every day to write tests for example, it's a massive time saving and I wouldn't want to write tests manually ever again.
Hey, I did not. As mcbetz mentioned, I tried vibe-coding and anything you see is generated by LLMs. He linked to my blog post where I summarized the journey.
ZenQuery looks nice, I like the focus on supporting regular files.
A database desktop client, built with Tauri & SvelteKit.
"But there are many already!" I hear the crowd exclaim.
I respond, "Yes, but..."
It's really something I want for myself. Lightweight, as fast as humanly possible, extensible via plugins (in fact the entire app is mostly plugins, with a small core to glue it together), and a tiny bit of LLM (call it AI if you wish) integration to ask questions about the database or generate/review queries.
Thank you, I appreciate the in-depth opinion! I think Tauri with SvelteKit is going to be my first prototype, likely followed by gpui because I'm impressed by Zed which is built with gpui.