After a few months of pretty heavy use, I can honestly tell you: I am amazed at what I can accomplish in a few hours. And I still don’t know what I don’t know.
Understanding what’s public and what’s private, those are the major gating factors for most IT organizations. And when I say most, I probably mean yours. I’m in the 98th or 99th percentile of hands-on AI usage among IT leaders right now, and I’m telling you: none of us are ready to unleash this on our corporate networks yet.
But that doesn’t mean it can’t deliver results.
In a private environment, a contained lab with meticulously scrubbed, anonymous data,AI can still do incredible work. Here’s a real example from my home lab:
I wanted to connect NetBox to my UniFi controller and pull live network data into my DCIM system. I asked Claude to build the script. It did. I ran it in an isolated environment and watched the data flow through.
Then I did what any responsible IT person should do: I’d built this in Proxmox. I’d taken snapshots of both VMs before running anything. I reset them both back to their default configs. And I said: “Hey Claude, do it again, but this time, document every step and every script you used so I can review them and do it manually.”
One or two iterations of troubleshooting in a safe, isolated environment. Then you take Claude’s scripts, review them with human eyes, and deploy them manually in your production environment.
Is it as fast as letting Claude do it directly? No. Can you hand a package of plaintext scripts to your security team to review before running them? Yes. Does this make you 100 times faster than building them yourself from scratch? Yes.
Is Claude 1,000 times faster if you just let it run? Also yes. Are you willing to take that risk? If the answer is yes, maybe you shouldn’t be doing this.
In the weeks I’ve been using AI as an assistant, it makes about as many mistakes as a human does. “Let me edit that page for you”, oops, it overwrote it instead of editing it. Just like I’d expect from a junior admin. “Can you fix it?” Claude checks: “Yes, I have access to edit it.” Or: “I ran out of context and don’t remember what I just wrote. We have to redo it from scratch.”
Circa 1995: “Did you save your work?” “No.” “Then why are you asking the IT guys if we can fix it?”
Some things never change.
There is real value in AI building processes for you. There is an unknown amount of risk in letting it execute them for you. The gap between those two statements is where the governance lives.
There will be more stories about this as I learn where the guardrails belong and what AI can and can’t do safely. We’re learning together, in a controlled environment.
I couldn’t imagine a better thing to bring to a table than “I have done it” versus “I think it will work.”
Until next time. Have a good night.