LLVM just dropped a new AI Tool Use Policy (commit: 18695b27d565). If you're not familiar with LLVM, it's the compiler infrastructure that turns your code into actual machine instructions. I've always loved LLVM - it's where software meets silicon.
The policy is straightforward: use whatever AI tools you want, but you have to actually understand what you're submitting. LLVM isn't "banning AI." They're doing the smarter, more practical thing: they're requiring basic professional responsibility. Which sounds obvious, and yet here we are.
The important bits:
First, LLVM is very clear about ownership. If you submit it, you own it. Not the model. Not the prompt. Not "the vibe." You're the author, and you're accountable.
That means you need to read and review all LLM-generated code or text before you ask maintainers to spend their time reviewing it. And if someone asks you questions in review, you need to be able to answer them. This is basically a new "RTFM, but for your own M" policy. Read and understand your own damn code. Wild concept.
They also want transparency. If a contribution was substantially AI-assisted, label it (they give an example trailer like Assisted-by:). This isn't about shaming people for using tools. It's about being honest with the humans who have to reason about the change later.
LLVM also draws a line around automation: no autonomous agents taking actions in LLVM spaces without human approval. That includes things like bots that comment, review, or otherwise create noise without a person explicitly driving.
And one detail I really like: don't use AI to knock out good first issue tickets. Those issues exist to help new humans learn the project. If someone wants to speedrun contribution badges, they can do it somewhere else.
LLVM also names a thing: "extractive contributions." That's when someone dumps a pile of unvetted output onto maintainers and calls it a "PR." The cost doesn't disappear. It just gets transferred to the people least interested in cleaning up your mess.
Why I'm on board with HITL
HITL isn't about being anti-AI. It's about not wasting other people's time with garbage you haven't bothered to understand yourself. Open source maintainers are basically the support team for the world, except they don't get paid and they can't escalate your ticket to "priority: never."
LLVM's policy is a polite way of saying: if you can't explain what you're submitting, don't submit it. And I'm a fan. Nice job y'all.
-Sethers
