Similarly, I was reconfiguring my home server and having Claude generate systemd units and timers was very handy. As you said you do need to know the material to fix the few mistakes and know what to ask for. But it can do the busywork of turning "I need this backup job to run once a week" into the .service and .timer file syntax for you to tweak instead of writing it from scratch.
I think it's just a turbo mode for figuring things out. Like posting to a forum and getting an answer immediately, without all those idiots asking you why you even want to do this, how software X is better than what you are using etc.
Obviously you should have enough technical knowledge to do a rough sanity check on the reply, as there's still a chance you get stupid shit out of it, but mostly it's really efficient for getting started with some tooling or programming language you're not familiar with. You can perfectly do without, it just takes longer. Plus You're not dependent on it to keep your stuff running once it's set up.
Claude and others are still in the adoption phase so the services are good, and not user hostile as they will be in the extraction phase. Hopefully by then some agreement on how to setup RAG systems for actual human constructed documentation for these systems will be way more accessible, and have good results with much smaller self hosted models. IMO, this is where I think/hope the LLMs value to the average person will land long term. Search, but better at understanding the query. Sadly, they will also been used for a lot of user hostile nonsense as well.
No. I've been a sysadmin before and know how to write the files from scratch. But Claude is like having a very fast intern I can tell to do the boring part for me and review the work, so it takes 30 seconds instead of 5 minutes.
But if I didn't know how to do it myself, it'd be useless- the subtle bugs Claude occasionally includes would be showstopper issues instead of a quick fix.
