>>1AI wins because its "Worse is better", it provides 'good enough' content that is orders of magnitudes faster/cheaper.
To win over, you need to construct entire alternative ecosystem(Math2.0, Software2.0).
I'd think the only viable alternative is open-source packages registries expanding to cover all possible use-cases,
(though the maintainers might not like it), where AI code would be outcompeted by pre-existing "packages" solving
a domain more precisely than vibe-slop crafted in one-shot.
It however requires expanding the scope of the idea of "package registries" to encompass orders of magnitude more
content,like a LEGO set growing its component-library to allow more construction options. This is of course requires immense focus on following user needs, which open-source devs famously ignore, so AI wins here purely by appeal to user concerns and of course re-using the same "package registry ecosystem" you built up.
So basically you're being outcompeted by this.
1.A user/company has unique need which has no app/library/package answering its domain of application.
2.AI crafts vibe-coded, "Worse is Better" v0.01 app that somewhat works(90% of time).
3.If a package/app/library existed that answered #1 need, the AI would reuse it or the user would find it better for their usecase, removing the need for #2.
4.Snailcats lose, since #3 was crafted with preferences of open-source devs which use-case diverge from 'average consumer' in the domain and have 10% of AI app user-friendliness.
5.AI app gets improved, by user yelling at Claude/GPT to fix all bugs. Snailcats review each pull request leisurely once per week and reject user concerns as "not really fit for our open-source design".