It still gets complicated if you have many interconnected devices to manage of course, but at least it limits the obscurity of the most complex devices in your network.
I'm open to alternatives though, if anyone has some to share.
I have found a couple of ways of doing documentation.
1. I document myself, with my style: I have a big Obsidian vault with work related documents, it's loosely based on Diataxis [0] method. I live document the part I'm working on once, and when I need to repeat the deed, I can update any changed parts, if any.
2. SaltStack. People doesn't prefer it much when it comes to Ansible, but pillars and recipes allows automation tons of stuff in a self documenting format. Plus, if you wish SaltStack has drift control and anchoring (like the default CFEngine behavior).
3. Terraform. Not incorporated by me, but a project uses it, and we use it in turn. What's nice about it is scaling. They have a variables file, and change the number of the server type you want to scale, and apply. It's scaled automatically. Infrastructure as code at best.
You want to installs with specific versions and packages? You can use xCAT for RedHat and Debian systems (which will be replaced by Lenovo Conflence soonish). You can re-provision a system to the exact state you want with three commands and, within 15 minutes, your system will be ready. Since it uses the repositories local to xCAT, no version drifting occurs.
I personally install Debian stable, enable security updates, and periodically make sure that things are working. Not taking much time (10 minutes per month?).
[0]: https://diataxis.fr/
For example, package management actions map to specialized packages per OS, and for that specific package, documentation gives how the specific package calls the binaries or use the libraries. After a certain point you start to build call trees and see everything in your mind before pressing enter.