LLMs allow a new lower-effort tier of prototyping.

Before, the way to prototype was just to do it rough and ready manually, with throw-away code.

But now there's an easier / cheaper tier before that point: simply ask an LLM to do it.

Use LLMs to duct tape something together, cache the result, and then later replace it with traditional code if you want to keep it and use it a lot.

In my Code Sprouts hobby project, I needed a blank javascript object that matched a given typescript schema.

I could have theoretically found a Typescript parser that could run in a browser and then do some kind of processing on the AST to generate an object.

But that's a lot of work and it's a simple hobby project.

Instead, I just asked an LLM to do it, and then cached the answer in a "compiled" version of a sprout.