Post

re: the open questions: I suspect the LICENSE of the original repo is most relevant here as it signals repo owner intent. Using an agent to clone a MIT repo fits the same spirit of code expansion as if a human did it. I would not recommend using an agent to clone a GPL repo.

Yeah, I also deliberately picked MIT as the license because that was previously used by both JustHTML an the html5ever Rust library Not sure how I should do the (c) line in the MIT license file though!

Can't speak to the legal requirements, but in terms of professional courtesy you should probably include the original authors, especially if it's a 1:1 rewrite in another lang. It feels equivalent to translating a book to a new lang - you wouldn't get sole copyright there?

I'm sharing this because it's quite a stark demonstration of what's possible using the latest frontier LLMs at the tail end of 2025 Is it a GOOD idea to churn out a 9,000 line direct port while watching Daniel Craig chew on the scenery? That's a whole other set of questions!

Impressive! I'm curious, the license says it's copyright you. Is that accurate? Who actually owns the IP, if an AI agent wrote all the code? IANAL and even less a court, so I don't really know, but it's also interesting to me, and it's likely central to some of the questions at the end of the post

That's one of the questions I should have included in the post, I'll add that now I don't know for certain. Most of my AI-assisted code involves WAY more of my own interactions - in this case the majority of the input was the code written by Emil for JustHTML itself

Do you know if someone has attempted the "meta challenge" of building a repo that will write an implementation in your favourite language? Would be interesting to see how well the idea "use LLM as compiler of specs to super personalised implementation" works today

I’ve been making a tool to follow along w/these things. has a web UI as it draws its decision graph of observations goals options outcomes & actions in a DAG & helps you see its process & guide things along. Disclaimer: I made it over nights/weekends the past 2wks. Would be curious your 2 cents

notactuallytreyanastasio.github.ioDeciduous - Decision Graph ToolingTrack every goal, decision, and outcome in AI-assisted development. Survive context loss. Query your reasoning. Never forget why you chose approach A over B.

@simonwillison.net Absolutely fascinating stuff Simon. I had the exact same idea a few days ago, ironically when I encountered a different parsing library I wanted to port to Swift for @plinky.app). I very much appreciated the moral questions at the end too, since similar thoughts crossed my mind.