On Tue, 17 Nov 2026, Kevin Wolf wrote:
Am 17.11.2026 um 21:40 hat Eric Blake geschrieben:
[Chiming in late]
On Wed, Sep 30, 2026 at 10:54:30AM +0200, Paolo Bonzini wrote:
Until now QEMU's code provenance policy declined any contribution
believed to include or derive from AI-generated content. A blanket ban
was easy to maintain while LLM output was rarely usable on its own, but
as the tools improved an absolute prohibition has become harder to
justify and enforce.
That ban is still impacting contributions today:
https://lists.gnu.org/archive/html/qemu-devel/2026-11/msg03382.html
https://lists.gnu.org/archive/html/qemu-devel/2026-11/msg03845.html
And at least those of Christian's patches that I merged are a good
example of how AI-assisted doesn't automatically mean bad quality. These
patches were great, some of them better than the average human-only
patch. I assume that he still spent significant time on these patches
and didn't just send out the first thing AI gave him - which is how AI
should be used.
The problem is not people using AI but people using AI who otherwise have
no idea what they are doing and send the slop unreviewed and untested (or
submit tickets that way) which will put all the work to understand and
test these patches and tickets on maintainers. This might be OK for simple
fixes, one liners or trivial patches but causes problem for larger
changes, So this policy is meant to make that clear to not do that and AI
usage is only allowed if the submitter does the work of reviewing and
testing the generated patches before submission and has enough
understanding to do that and answer review comments themselves. As people
who are sending slop probably won't read this policy either I hope the AIs
will follow AGENTS.md and tell them.
I've also come across another project where the AGENTS.md said agents
should not create any tickets, humans should do that by hand (which
probably filters the fire and forget people at least) and that AI should
prefix generated patch subjects with [AI] (we have AI-used-for tag here
that's similar) and also to prefix any generated text with a robot emoji.
That's probably overkill but the point is to tell AI in AGENTS.md to mark
generated content itself so it's easy to see at review. Just mentioning
it what I saw others doing who don't want unreviewed AI content, not
proposing we do the same. But maybe putting in AGENTS.md that tickets
submitted by agents should have an AI tag in title or somewhere easily
filterable might help to put them at the end of the queue in triage?
Regards,
BALATON Zoltan