Field Manual · Ed. 2026

AI systems, built to stay running.

Independent operator for AI systems, automation, and infrastructure. Research pipelines and internal tools built to keep working.

See the systems → Get in touch

Speed Research pipeline
30 min → 50 sec

Workflow 8-agent pipeline
with QA gate

Runtime Used daily
Running 24/7

Stack Local LLMs
zero cloud for agents

§01 · Practice

The work is operational, not theoretical.

AI and automation lead. Infrastructure keeps the systems reliable. FPV stays secondary.

A · AI

Multi-agent systems, research pipelines, and internal tools.

Production workflows with explicit handoff, review gates, and monitoring. Local LLMs on owned hardware where it matters. See the systems →

B · Infra

Infrastructure experience carried into software.

10+ years running Active Directory, Windows Server, and enterprise automation. That instinct for durability leaks into every build. About the operator →

C · FPV

FPV as a secondary practice.

Freestyle and long range, four custom builds, DJI O4 Pro on every quad. Not the lead identity, but part of the same operator habit. See the fleet →

§02 · Selected systems

Running in production, not demos.

A current slice. More on the Systems page.

Active

Multi-agent AI organisation

A production multi-agent system with an explicit org chart — orchestrator, infra, QA, Windows, and integration agents — handling research, review, content, and monitoring. Runs on a home Ollama server.

Research: 30 min → 50 sec. 8 agents, 3 LLMs, zero cloud compute for the agent lane.

OpenClaw Claude Sonnet Ollama Python

Lane: AI · Infra
Stack: Local-first

Active

Discord shopping list bot

Discord-native list with interactive buttons, store-aisle sorting, quick-add, and daily price alerts. Flask + Nextcloud. Used daily by the household.

Items auto-sorted by Norwegian store walkthrough order. Price check runs every morning at 08:00.

Python / Flask Discord API Nextcloud Kassalapp

Lane: Automation
Users: Family · daily

Full systems index →

§03 · Method

Build it like it has to survive contact with operators.

Best fit: a real workflow, a real owner, and a cost when it breaks.

01 · Scope

Find the real constraint.

Where is time, attention, or reliability actually leaking? Automate that, not the nearest ritual.

02 · Build

Keep the system understandable.

Boring, durable parts where possible. Clear prompts, sane tooling, enough structure that someone else can operate it later.

03 · Run

Ship with verification.

Review gates, monitoring, or human checkpoints — so the result is useful in production, not impressive for five minutes.

§04 · Engage

Have an AI system that needs to outlast the demo?

Most useful on practical work: automation, research pipelines, internal tools, and infrastructure-minded AI integration.

AI tooling and workflow automation
Infrastructure-minded system design
Technical builds with real owners

Start a conversation More examples