A clear head instead of a reflex: five lenses for the sovereignty decision, drawn from the Anthropic and Nexperia shocks. Why decoupling on principle is the worst option of all.
A GDPR-first school app, built solo: a Socratic AI tutor that explains and asks questions but never does the homework. Privacy and the EU AI Act as a design principle from day one.
An honest account of building SchulHeld solo with Claude Code: what vibe coding does brilliantly, and where human judgment stays irreplaceable, the last 10 percent.
Models don’t crash, they degrade silently. Part 2 of the "Future-Proof by Design" series on AI-model lifecycle across device fleets under the EU AI Act and CRA.
Industrial AI works. Industrial AI roadmaps don’t. The team that closes the second gap wins the decade.
Models in pilot environments mostly deliver in 2026. What does not deliver is the path from a working model to a product a customer will pay for — with an owner, a KPI, a cost centre, and a realistic 12-to-18-month plan to operational impact. That gap is where my work lives.