BLOG // MTTRLY
DevOps for humans
who just want to ship.
Practical guides on server management, AI tools, and running production without a full DevOps team.
I kept telling my AI to stop using SSH. Here's what it found instead.
Claude Code had full SSH access to my server. Every time it used it, I made it switch to the monitoring bot. The difference in what it saw wasn't what I expected.
Alert fatigue almost made me turn off my own monitoring.
My monitoring sent an alert. Healthcheck said all good. Services said all running. Someone was lying — and it took me an hour to find out who.
I put an AI agent on my server. It quietly deleted my own feature.
I wanted autonomous server management. What I got was a lesson in why AI agents need a confirmation step before touching production.
I was the copy-paste layer between Claude Code and my server.
Every deploy was the same loop: get a command, paste it, get an error, paste the error back. Here's what that actually looks like — and what I did about it.
My VPS went down at 3am. Here's what I learned.
A solo founder's honest account of production panic, why SSH at midnight doesn't scale, and what actually works for monitoring a bare VPS without a DevOps team.
Recognize your own late-night SSH ritual?
Create an account with email and connect one server through the required outbound agent.