It's quickly becoming a replacement for Google for me. When I want to know more about a topic I don't usually go to Google or Wikpedia as often, instead I'll start a dialog with Claude about the topic.

I also use it for writing, exploring complex topics or as a sounding board for interpersonal issues.

I'm fully managing multiple NixOS installs via CC. I set out to do it just as an experiment, but it's been working great and I've expanded usage. When I switched my desktop from Windows back to NixOS last night, the first thing I did was enable SSH and setup keys so my NAS can access my desktop and Claude Code has made every single configuration.nix change for me. I also use Claude Code to orchestrate my local LLM setup. It makes recommendation on local models and facilitates downloads when provided a huggingface API token.

My next experiments will entail using Claude Code to bootstrap some local LLM agents exposed as CC commands so not all of my system work will need to round trip to claude.ai to be completed.

For translations. All SOTA LLMs are good at mainstream languages these days, but when you need more niche stuff, some models can be better at some specific thing than others. Claude Opus in particular seems to be the best at Lojban, for example.

Yes! I'd say probably more than 1/2 my tokens are unrelated to code.

My favorite is I had (and still do have) a whole conversion about the water in my pool. I send it pictures of my water and test strips and it suggests how much of which chemical to add.

I asked about a recipe.

I used it to translate handwritten German from my grandmother.

I brainstorm business process ideas with it.

I ask it for medical advice (like, what should I google to find out what this bump is)

I brainstorm product ideas with it, like a PM.

And that's all just in the last three weeks.

Yes, of course. It’s good enough as sparring partner in thinking, e.g. when outlining product strategy or preparing copy. Of course if you know what to ask for or how to spot hallucinations.

Yep.

I use it for feedback on things I've written.

It's not as good as a good editor who understands what you're writing about.

But it is so fast and it really does help.

I know people that do (typically mathematics) since they pay for Claude Code anyway. I often tell them that they shouldn't.

0.1% of the time i ask questions, usually not about politics. since i dont expect much from it there

mostly sys admin things but yeah

That's besides the point, isn't it? There is a high likelihood that these models, these companies, and the people building them are going to be central in shaping future conversations and thought. Why does it matter what they're used for right now?