Here’s how it is:
Two months ago I wrote about teaching Claude to write Pony. That post was about the process — mentoring an LLM like a junior developer, building up engineering principles, establishing review loops. Today, I’m here to talk about the LLM skills I’ve been working on. They’re in ponylang/llm-skills, and if you’re working on Pony with an LLM, you should try them.
In my last post, I mentioned that the new json package uses persistent data structures under the hood. Every JsonObject is backed by a CHAMP Map. Every JsonArray is backed by a HAMT Vec. I glossed over why, and I want to fix that.
I’m not really sure how to tell the story of me teaching Claude to write Pony, so I’m just going to tell it and see how it goes.
A year ago, every LLM I tested on Pony produced the same thing: a weird Python/Pony hybrid that didn’t compile and didn’t understand Pony’s semantics. I’d ask for a program that prompts you for your name and prints “Hello {name}.” Simple stuff. They all failed miserably. So when I sat down two weeks ago to try again with Claude Code, my expectations were modest.