[scuff3d]

That's super interesting to hear. I see people say all the time "oh I would never use it in a big project because it's changing so fast", but then see awesome projects like Ghostty and Tiger Beatle written in it, so clearly they're doing something right.

I'm surprised to hear LLMs have been helpful. What little I tried to use them for with Zig, all the information was way too out of date. But I was mainly just doing research, not running an agent at the time, so that's probably why.

[mitchellh]

Yes, LLMs with Zig are absolutely useless without agentic behaviors. Throw an agent in the mix and it's totally fine. Bonus points if you pair it with some basic agent skills so it knows how to look up language references, stdlib files, etc. then it does even better.

[scuff3d]

I'll have to give it another try the next time I play around with Zig. I decided to let all the big changes with IO and what not play out before diving back in lol.

I've been slow on the agent pickup. The ethical issues really bother me (LLMs in general), though it's clear I'm not gonna have a choice in this industry so I've been caving recently. The persistent narrative that agents can build entire complex software systems is also irritating, because that clearly doesn't seem to align with reality.

[eevmanu]

Would you mind sharing or pointing me to any links that explain how you've set up Zig to work effectively with LLMs using agentic features?

I was thinking that downloading the full official documentation, separated by sections inside the repository someone is working on with Zig, could be useful, but maybe there are more optimal ways to approach this.

Thanks.