Mastodon Feed
Boosted by aredridel@kolektiva.social ("Mx. Aria Stewart"):
cate@hachyderm.io wrote:
It's a hard time to be an OSS maintainer. I see why many have taken a no-AI PRs stance. In general the worst part of AI is spending more time to fix something than someone spent "doing" it - whether that's a document, a PR, or a process.
Steve Yegge's "Vibe Maintainer" piece proposes an alternative: using AI agents to triage ~50 PRs a day, fixing contributors' code yourself, cherry-picking the good parts, making rejection the last resort rather than the first.
It's interesting and worth a read. I don't think no-AI PRs is a sustainable strategy, and - as Yegge notes - there are real downsides to "just fork it".
But the token cost to do this at scale seems very high. This is my continual complaint about many of the most bleeding edge AI workflows - how much do they cost, and how well does that cost translate to value?
In this case if the maintainer is going to spend tokens reconstructing your contribution anyway, the highest-value thing a contributor can do is write a really good issue. Describe the problem clearly, show your reasoning, let the maintainer's agent do the implementation.
In this time, being able to produce the specification seems higher value than producing the artifact. That's harder and requires more critical thinking. But with output abundant, critical thinking is now the scarce resource.