I'm an ambassador for Agile Testing Days this year, which mostly means I get to give away a ticket. Potsdam, 16-19 November 2026. To enter, answer one question. That's it. The prize is an on-site co...

The same cold, secular countries keep topping both the metal and the software testing maps. A half-serious theory about trust, rebellion, and how much a place lets you leave unsaid.

Seven years after the Mountain of Test Automation Doom, the AI version looks more like a forest than a mountain. What managers miss when they ask their teams to 'just use AI'.

I wanted to know which files in my knowledge base actually get used together, even when they share no words. Twelve weeks of behavioral observation later, the graph drew its own clusters.

Stacking validators on validators doesn't make a probabilistic system deterministic. Match the architecture to the risk appetite, and the AI governance conversation stops being binary.

I spent four years building trust with AI assistants. A month ago I went all in on Claude. Then it ignored every instruction I'd written. The models I trusted broke first.

I mounted a FUSE filesystem to observe how my knowledge base documents are used together. Day one: 690,000 events. Most of them were noise. This is the detective story of finding the signal.

My friend Patrick Prill mapped eight ways organisations lose sovereign agency to AI vendors. I read it and recognised the risks I'd been designing around for over a year.

The knowledge base could find Fredrik in 85 documents. It just couldn't tell a coherent story about him. This is the story of data quality, entities, and why 'just store everything' breaks down.

How I evolved my knowledge base search through five phases, each driven by a concrete failure, from pure vector search to hybrid retrieval with cross-encoder reranking.

I built an AI assistant that can't do anything without asking first. That's not a limitation. That's the design.

I built a knowledge base that made me dramatically faster. Then I couldn't stop. So I built a daily planner to fight back.

How I built a local knowledge base that indexes my emails, calendar, notes, and conversations — and why context infrastructure matters more than the model.

Vibe coding implies something passive. Like the hard part is the prompting. I built a knowledge base, a testing tool, and a training pipeline with AI. None of that was vibes.