Like, I prefer to have my own snippets and set them up the way I want, but hey - the convenience of LLMs... idk
For example, if I want to add an image to a document, the necessary full command to type this might be `\includegraphics[scale=0.5]{Image_file_name}`, which is a lot to type for image-heavy documents.
But with UltiSnips, I can write any shortcut (e.g. by typing `img` and pressing the `Tab` key or alternatively letting it auto-complete depending on my snippet setting), then the whole command is typed out: `\includegraphics[scale=0.5]{$1}$0`. My cursor would immediately be placed at the $1 position, and pressing the `Tab` key would take me outside to the $0 position, so I can keep typing.
It's simple and predictable for myself as a user to easily write the full command, whereas I predict that a solution that uses an large language model (LLM) for this context may add complexity without making the task easier for my purposes (though LLM-based software may be more useful in other contexts).
Also, why wouldn't you just use both? Snippets let you quickly use known patterns, LLMs let you quickly write the more generic case of boilerplate that doesn't align well with snippet use.