I've spent some time on learning if an LLM can produce Ansible playbooks for automating *basic* VM creation for the European cloud provider Scaleway: it can't.
The inference keeps mixing current and outdated information, and mixes in fixes for non-matching scenarios: AI slop.
It worked very well for what's common with American companies, like AWS, in the basic Ansible provisioning use-case.
One must read the docs and learn IT by doing it like the pros. It's an art.
@tg9541 Your first sentence almost sounds like a parody. The rest of it sounds exactly like my experience trying to get it to help me configure ollama and openclaw to work together on different computers today. It has no fucking clue what it's doing.
@davidvasandani How do you like OpenClaw so far? I haven't gotten it to work but I'm not really convinced it's worth it. Not sure where all the hype is coming from.
@apodoxus It’s incredible and is a solid example of how I think we’ll interact with personal assistants in the future. I’m still experimenting with what’s configurable within openclaw vs how you can define an agent in prompts via the BOOTSTRAP, SOUL, IDENTITY, etc docs.
@davidvasandani What do you do with it that you couldn't do before?
@apodoxus Nothing. It just does it for me.
@davidvasandani Google, ChatGPT, Brave, DeepSeek.
@davidvasandani I used Gemini. There was nothing wrong using it to create Ansible playbooks for configuring VMs. It even navigated me through working with IPv6-only nodes.
Where the problem started was when I tried using Ansible for creating Scaleway VMs. That's possible but the LLM couldn't get it done.
Maybe I shouldn't use Ansible but the Scaleway command line client, or another API.
@davidvasandani no, I didn't try that. For VM creation with Ansible there are very few official docs to begin with.
@apodoxus
