I wonder if AI will eventually make faster languages (eg Rust, Crystal) nearly as simple to ship with, and therefore a better choice. Or maybe that will end up being a new language entirely.

So much of what goes into a language choice involves things that aren't the language itself but the ecosystem around it. Documentation, standard library, third-party libraries, build and deploying tooling, and so on.

So a huge part of getting people to move from an existing tool they know with a well established ecosystem to a new tool that perhaps doesn't have that yet is getting those things in place (and then getting people to want to use them). I can imagine that being able to boostrap part of that process could help with the adoption of new tools.

The counter argument is that the friction behind these things is actually a benefit, as languages are only useful when they're shared and it doesn't make sense to have a huge range of languages that all mostly function the same but are just different synatcically. So it's a feature that it takes time for newer languages to gain traction, because it's essentially being tested for suitability and usefulness by a slowly growing number of people, at first the innovators, early adopters and early majority on the curve and later on the ones that perhaps can't afford to spend time on something that might not gen enough traction to last.

On a related note, I'm curious to know what the impact of generative AI on language-interoperability will be. Every time I see a suggestion that WASM and Interface Types is the thing that will finally allow me to use that weird but useful Python library in my Rust project, I always feel like unless there's a performance consideration involved I'd really rather just have everything in the same language, so the idea of automated translation of libraries at the source level seems like it could be useful. No need for language interop when you can quickly put everything into the same language.