[Biopython] Generative AI policy for contributions to Biopython

[Peter Cock]

Replying to Ivan in-thread (they get the mailing list digest):
https://mailman.open-bio.org/pipermail/biopython/2026-April/017117.html

I personally would also be very reluctant to move to using AI for
reviewing code contributions.

Mitchell's Vouch https://github.com/mitchellh/vouch is evidentally
working well for them on Ghostty and other projects. He's a good
example of an experienced high profile developer apparently able to
use the AI assistance to great effect (I don't have the expertise to
judge Zig code), so that makes dealing with AI assisted contributions
to his open source projects an interesting test case.

Another interesting datapoint, Daniel Stenberg of the curl library has
complained about slop PRs in the past and the incentives that drove
this. He now acknowledges a significant improvement in the quality of
the AI submissions - which he attributes to recent (circa Spring 2026)
improvements in the tooling. eg
https://mastodon.social/@bagder/116464616490905282

Peter

On Fri, Apr 24, 2026 at 7:24 PM Ivan Erill <ivan.erill at gmail.com> wrote:
>
> Dear Peter,
>
> I don't think there is a viable answer that does not go in the direction you point towards (i.e. some sort of ban), if the aim is to keep manually reviewing Biopython. It simply does not scale up. I believe the discussion now for open-source projects like Biopython is whether to maintain manual review or fully embrace AI both as contributor and maintainer/reviewer. I, for one, would rather stick to manual review.
>
> How to maintain manual review in an AI world, as Markus points out, is an entirely different question. Some initiatives, like Vouch (https://github.com/mitchellh/vouch) explore the old concept of a web-of-trust. Simply pre-authorizing/vetting contributors via an old fashioned mail list, rather than through github, may be an effective way to screen out fully-automated PRs or PRs intended only to boost creds. It's essentially a throttle question. Constrain too much and you'll lose some quality contributions, open up too much and you'll get swamped. A fine line to walk.
>
> Incidentally, thank you and the main reviewer cadre for keeping Biopython alive and kicking.
>
> Ivan