Henry. AI systems, FPV, and the IT underneath.
The background behind the notebook: who I am, what I run, and how I work.
I'm Henry, based in Bergen. I come from ten years in IT infrastructure. Most of my working attention now goes to AI agent systems: building them, running them on my own hardware, and writing down what actually happens.
The AI work is operational, not demos. A multi-agent organisation runs research, review, and monitoring on local models around the clock. An operator stack routes daily work through delegation, persistent memory, and verification.
IT is the layer underneath. Microsoft 365 and Azure administration, identity and access, endpoint management, virtualisation, networking, monitoring, and automation in PowerShell, Python, and Ansible. That work built the habits the AI systems depend on: reliability, recovery, permissions, and the boring parts that decide whether something survives real use.
FPV is the counterweight. It is personal, visual, and physical, and most of the footage on this site is mine.
Evaluating how I work? Start with the systems index for what is running and what state it is in. The notes are the failure reports: what broke, why the checks missed it, and what changed.
No agent's claim of success counts until a check that could have failed says so. From the day 41 of 41 tests passed and three of them were failing. Read the note →
Advice first, mutation second, verification last. One working path for every change, whether a person or an agent makes it. Read the note →
Generation creates artifacts. Operators decide what ships. Nothing an AI tool produces is applied by default. Read the note →
If every fix is heroic, the system is badly instrumented. Most operations work should be boring. Read the note →
Behind the rules: decisions get a numbered record, incidents get a runbook with the root cause, and a backup does not count until a restore from it has been tested.