AI, organisations, and responsible change.
The AI Act became applicable six days after an Omnibus changed part of its timetable. Some obligations arrived as planned, while others moved years into the future. I think that split exposes a tension built into the Act itself.
Moving professional work into a company ChatGPT workspace looks like a migration. The harder question is whether anything arrives that outlives the person who put it there.
The new mark for this blog began in the earth's shadow. Building it with Claude turned a private connection into a working identity, and gave the writing somewhere to go again.
Large language models produce fluent language in seconds. Piaget helps explain why understanding still has to be constructed.
Quantum computing still feels distant, yet anyone can already write, simulate and run simple quantum programs. The software ecosystem is arriving before the hardware matures.
ChatGPT Work and Kimi K3 made me look beyond the engine. Increasingly, the meaningful product may be the environment in which a model can actually work.
Over the past few weeks I've found myself reading widely differing takes on AI. Some see the biggest speculative bubble in history. Others see the foundations of a new industrial era. Perhaps they're simply looking at different parts of the same landscape.
Why do I keep choosing proprietary products when open-weight models promise more control, flexibility and independence? The answer may lie less in principle than in what is actually usable, affordable and reliable today.
A Dutch quantum computer just reached its first customer, not for its power to solve problems yet, but to learn what building on this new kind of computer means.
Compute is necessary but not sufficient. A LinkedIn debate sharpened an argument about European AI infrastructure into a longer chain: energy to demand.
VOLT's proposed Rotterdam AI campus would be among Europe's largest, if built. The real scale, and why Dutch soil isn't the same as durable Dutch control.
A vocal minority told me I wasn't entitled to use a frontier model for a "simple" diagram. It reveals a split running through centuries of regulation: license the person, or license the maker.
Fable is the best model I have, and the heaviest. It drains my usage faster and costs more past the limit. On the price of reaching for the strong tool.
Vanta makes vendor trust continuous instead of episodic. What it automates is real. What it still can't do is tell you whether that's enough.
A vendor I vouched for got challenged twice in one week. I couldn't answer with more than trust, so I built a method to climb past it.