LLMs are generative and do not have a fixed output in the way past autocompletes have. I know when I accept "intellisense" or whatever editor tools are provided to me, it's using a known-set of completions that are valid. LLMs often hallucinate and you have to double-check everything they output.