Most network outages I have seen were not caused by hardware dying. They were caused by somebody changing something on a Friday afternoon without a way back.
I have been rebuilding and cleaning up networks for a long time, from a parish campus with cameras, door access and livestream gear to my own shop. The pattern is always the same. The work is not the hard part. The hard part is making change boring.
Make change boring
- Write down what exists before you touch it. A current inventory and a backup of every controller and device config. Not the plan. The truth.
- Segment first. Cameras, door access, guests, office and media storage each get their own VLAN. A mistake in one place stays in that place.
- Change one thing at a time, in a window, with a rollback. If you cannot describe how to undo it in one sentence, you are not ready.
- Test from the user’s side. Green lights on a switch do not mean the livestream works. Run the actual service.
- Log it. Date, what changed, why, and who approved. Future you is a stranger who needs the notes.
Where AI fits
I use AI for the parts that are tedious and error-prone: comparing a live config against the intended one, turning a pile of device exports into a plain-English inventory, drafting the change plan and the rollback steps, and summarizing logs after a change goes in. It prepares. I review, approve and push. Nothing gets applied to production because a model said so.
What you get
Faster changes because the prep is done. Safer changes because every one has an exit. And a network you can hand to someone else without a long phone call. That is the whole pitch: fewer surprises, and a clear paper trail when there is one.
If your network has grown by accident, let’s talk about a clean second start.