There are many barriers to adoption for a new programming language looking to go mainstream. You have to attract a small group of enthusiasts, build an ecosystem of high quality libraries, help new people learn, and eventually grow a talent marketplace.
The rise of LLMs adds another wrinkle—if programming with an AI assistant becomes commonplace, the degree to which the AI is good at the programming language could be a significant benefit (or detractor).
An LLM-first programming language might leverage AI assistants to make adoption significantly easier. I love the rust compiler because it is basically a self contained help system, but it requires study and practice to be productive. Imagine if you could become productive immediately and then learn more.
Not only that, an LLM-first programming language might have a deep repertoire of libraries, patterns, and problems it draws from that were specifically trained for language. I imagine this would be much more effective in day to day coding tasks than the aggregate output of blog posts that happen to be published on the internet and pulled into a blob of training data. It might even provide better context for other LLM applications to give more meaningful help (surely a compiler has some more precise data like the AST, the change, the types, etc that would be useful for figuring out what you are trying to do).