What I'm focused on now.

The main thing consuming my engineering time. Whiteaway runs a large e-commerce operation across the Nordics, and the fulfillment side (orders, warehouse picks, carrier handoffs, delivery confirmations) has historically been tightly coupled to a legacy Navision ERP.

We're taking that apart. The goal is an event-driven fulfillment pipeline where each domain (inventory, picking, shipping, returns) owns its own state and communicates through events rather than synchronous ERP calls, replacing brittle point-to-point integrations with a durable event backbone.

In practice that means a lot of domain modelling, finding the real bounded contexts in a system never designed with them in mind, and incremental migration so orders keep moving while we rebuild the rails under them. It's now split across several 2–3 person teams, and a growing part of my role is breaking business capabilities into scoped, dependency-aware work each team can run with.

Current focus: event schema design and the first domain extraction, getting order lifecycle events flowing independently of the ERP write path, plus the team breakdown for the next wave.

I'm responsible for rolling agentic development out across the engineering department. Not just picking a tool, but building the practices around it: how to scope work so an agent can run it well, how to review agent-written changes properly, and where the approach pays off versus where it doesn't.

That practice now lives in a Claude Code plugin I built and maintain: skills for planning a feature, running a whole epic with parallel implementer agents, verifying a change with real evidence, fixing CI and investigating production, plus a reviewer agent that reads each PR with none of the author's context. A guard hook stops agents from force-pushing, skipping tests or touching cloud resources without asking, and every skill change has to pass an LLM-graded eval suite. It runs in Claude Code, Copilot CLI and Codex, so nobody has to switch tools.

My own output is the test case: 680+ merged PRs across roughly 50 services since April, all of it worked agent-first.

Current focus: piloting the plugin in my own team, feeding what real use shows back into the skills, then taking it to the rest of the department.

A bicycle component wear tracker. I forget when I last replaced a chain or swapped brake pads, and I want the app to tell me before the bike does in a less pleasant way.

Rather than just counting kilometres, it pulls rides from Strava and Intervals.icu and cross-references them with the weather: a wet or snowy ride wears a drivetrain much faster than a dry one. Rust API with sqlx compile-time checked queries, so SQL errors surface at cargo build, not at runtime.

Live at drivetrain.thaudal.com.

A week-long challenge: a Tamagotchi for my wife. Hammi is a pet hamster that lives on her phone. It hatches from an egg, grows up over four days and gets hungry, bored and sleepy in real time, with push notifications when it needs something. The whole household shares the same Hammi.

The one hard rule was that it stays gentle: neglect only makes Hammi sad, never sick, and it cheers up as soon as someone looks after it.

An iOS app that syncs workouts across HealthKit, Strava and Intervals.icu, deduplicating as it goes so the same run isn't logged three times.

On the backburner while the Whiteaway work is intense, but maintained.