Current llms are delusional, and based on reddit bullshit. You probably have like 50% of the real speed on those vms because of "default" or "commonly belived" settings. Use a baseball bat on the agent, you have to hit it hard (yelling on it helps too), then you may get some useful info from it like setting big block sizes on ZFS and a correct arc/arc2 allocation and 31337 other settings which you never even heard of (neither did your agent)

Just feed it your CI playbooks; it'll know what you're trying to do. Tell it to never look at secrets (it will) and to make no mistakes (it will). 🤣 Why would anyone want to vibe code a homelab anyway? That's like paying for test answers to get your medical license; nothing good comes out of it

it's LLM assisted system administration.

Plenty to do with a homelab, don't get me wrong; LLM assistance is great. I'd be careful about letting a clanker bootstrap the whole thing though.

I think one of the big advantages of these robots slaves is they might take some of the maintenance burden of self-hosting

it's really just delegating and expanding cognitive ability. i've been running similar experiments, with great success. I already know how to do it, but in training an army of slaves with precise instructions, they do it for me. amazing times

Hopefully. I might still be using Nextcloud if I had something to fix it after every fucking update

imma sayin'

i doubt the "expanding" part on a large scale we're going to see decline in cognitive ability but for those of us who can actually use the tools, it will be helpful

I'll build my apps from source if I can just tell my guy to spin up a VM and clone the repo 💯

you're probably right for most people, although I hope that you're not. but for those who can leverage the tech.. godlike.

Script the change you want to see in the world