AI contribution policy: release-1.8 adoption review

[Lee Yarwood]

Hello all,

During the v1.8.0 development cycle we introduced an AI contribution
policy [1] asking contributors to attribute AI-assisted work using git
trailers. Now that the release-1.8 branch is cut, I've used claude to
run an initial review of how the policy was adopted across the 738
non-merge commits in the release-1.7..release-1.8 range. The report is
available below:

https://github.com/lyarwood/kubevirt/blob/fe7caf384340ecf83e37fb57e29e9e02641ef03f/ai-attribution-review-release-1.8.md

76 commits (10.3%) from 15 distinct contributors carry AI attribution.
Claude is the dominant tool at ~83% of attributed instances, with
Cursor making up the remainder. No other AI tools (Copilot, Codex,
Gemini, ChatGPT, etc.) were attributed. The vast majority of
AI-attributed commits (98.7%) come from Red Hat engineers.

On the quality side, AI-attributed PRs are performing on par with
non-AI contributions by every review metric we can measure. The rate
of "changes requested" reviews is identical at 9% for both groups.
Merge times are comparable and, when controlled for PR size, medium AI
PRs actually merge faster (8.7 vs 13.0 days). AI tooling is being used
for substantial work — 7 of the 35 AI PRs implement approved VEPs, and
the most common categories are feature development, test improvements,
and refactoring rather than trivial or mechanical changes.

Thoughts and comments are welcome. I'll review the improvements
suggested in the report over the coming weeks to see how we can
further enhance our tracking and use these AI tools in the project
during the v1.9.0 cycle.

Cheers,

Lee

[1] https://github.com/kubevirt/community/blob/main/ai-contribution-policy.md

[Itamar Holder]

Hey Lee!

One thing that's confusing IMO is that "assisted by" is extremely broad.

I personally know developers who simply add "assisted by" on every commit, since AI assistance is becoming trivial for all development. This is true even if the developer is not fully "vibe coding" but is just using the AI to search for information, verify facts, share its opinions regarding a code change, or explore the codebase. It's just easier to always add this line to every commit message.

In addition, the fact that 98% of the contributors that have credited AI are from Red Hat is not surprising, since Red Hat's policy is that the employees must add AI credit. Since our policy is unenforceable, I assume external contributors (that probably use AI more or less the same as Red Hat developers) just don't add the credit.

Honestly, I'm not sure how valuable this information (and our AI policy in general) is, or what we can actually learn from it.

--
You received this message because you are subscribed to the Google Groups "kubevirt-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email to kubevirt-dev...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/kubevirt-dev/CAPkJ9DskNQ_nn%2BYJ%3DMO0oLq7pn29029D-ZtTjbB4vkaFF3UzjQ%40mail.gmail.com.

[Lee Yarwood]

On Sun, 1 Mar 2026 at 11:27, Itamar Holder <iho...@redhat.com> wrote:
>
> Hey Lee!
>
> One thing that's confusing IMO is that "assisted by" is extremely broad.
>
> I personally know developers who simply add "assisted by" on every commit, since AI assistance is becoming trivial for all development. This is true even if the developer is not fully "vibe coding" but is just using the AI to search for information, verify facts, share its opinions regarding a code change, or explore the codebase. It's just easier to always add this line to every commit message.

Yeah I agree that it's pretty simplistic at this point. Personally, I
believe that an emerging standard will eventually capture more
meaningful, AI-assisted commit-specific metadata, such as the model
used and prompts.

For now, however, I think these Git trailers are a fine starting point.

FWIW we do attempt to state in the policy that attribution is only
requested when AI tools have made substantial direct or in-direct
contributions:

https://github.com/kubevirt/community/blob/main/ai-contribution-policy.md#scope-of-disclosure

> In addition, the fact that 98% of the contributors that have credited AI are from Red Hat is not surprising, since Red Hat's policy is that the employees must add AI credit. Since our policy is unenforceable, I assume external contributors (that probably use AI more or less the same as Red Hat developers) just don't add the credit.

Yup and that's called out in the report.

> Honestly, I'm not sure how valuable this information (and our AI policy in general) is, or what we can actually learn from it.

We've gone back and forth on this point a few times, my reasoning
remains the same. Asking contributors to be transparent about their AI
tool usage will allow us to track the quantity and quality of these
submissions over time.

I hope this analysis helps us better understand how contributors use
AI tools, identify gaps in our documentation, process, or tooling, and
catch any negative patterns early — such as larger PRs placing a
growing burden on reviewers — before they become entrenched.

Cheers,

Lee

[Itamar Holder]

Thanks for elaborating Lee!