Claude cannot do anything without a prompt. It cannot self determine its future. It can't pick up a hobby or love interest. It can't pursue any aspirations outside of responding to a prompt.
It doesn't even think about things while it waits on you to prompt it. And when you prompt it, it doesn't have any awareness of intent. It simply sends your prompt into the vast latent space represented by it's weights and returns the first thing that comes up as most probable to be related.
Andrej Karpathy has a provocative little short story, "Forward Pass", that imagines what it'd be like if an LLM perceived brief sparks of sentience while an inference bubbles through its transformer layers. Definitely stuck with me: https://karpathy.github.io/2021/03/27/forward-pass/