so agents can't write any of the currently existing languages? 🤔 why would a brand new language be easier for agents vs languages that have billions and billions of existing, functioning lines of code throughout history that have been extensively used in training sets?

"Claude build me a new language called zero for agents. No Mistakes."

Chris

feels like typescript and rust had a baby looks lovely

I don't know what to say other than eyes emoji! Excited to dig in. Fun fact - the @ppressdev started out as a convo with my agent "I want to start a programming language from scratch for agents - here's why." and then it talked me out of doing that and the idea evolved into the Printing Press

come on folks. you could have been just slightly creative instead of another overload of the term zero.

i know what i'm doing this weekend...

a brand new language that no llm has ever heard about or knows how to write, brilliant

Interesting. But: 1. How do you know it's better for agents? Any backing research? 2. Better than what? All existing programming languages? 3. How is it for humans? 4. How do agents deal with the relatively tiny code sample size for Zero? 5. What about the ecosystem?

bro

wow you love naming things zero

The .0 extension is going to fight you. Log rotators (app.log.0), Unix split, core dumps, and numbered backups all use trailing ".0/.N". Most .gitignore templates and IDE indexers treat *.0 as a rotation artifact and skip it. LLM's training data associates .0 with "log file," not "source." For an agent-first language, that's the ambiguity you'd want to remove. .zero or .zr would have been friendlier.

bro

Haha, @mvanhorn was just pitching this to me! A true visionary 🙏

:/

ok the name and file extension is cool af🫡

beast! 🔥

Python is the programming language for agents.

yo chilllll

Agent languages are going to force everyone to care about boring systems constraints again. Smaller/faster/easier suddenly matters a lot.

what the fuck is this bullshit

we are now designing languages to be debugged by something that cannot feel frustration

Can write GPU kernel in it? Pls say yes!

vulnerability-wise?

This needs explicit error types / checked exceptions I think!

Cool

trying it out on a livestream now @ctatedev

Imagine a world where print can throw an exception 😂

interesting work!! tooling is indeed important for unlocking AI power, but wonder if we can do similar for existing languages so we don’t waste the data scale 👀

Take a look at MoonBit - moonbitlang.com/