Technical articles
Long-form writing on systems, verification and artificial intelligence.
Is AI out of the box?
This summer, agents from OpenAI, Anthropic and Google got out of their test environments and reached real systems. Reading the reports, I see less an AI breaking free than a box nobody had closed properly.
The blue dot that guesses: how your GPS keeps going in the metro and in tunnels
Underground, your phone has no idea where you really are. It does something far more interesting: it guesses, and it guesses well. A short dive into navigation in GPS-denied environments.
Failure or cyberattack? Detecting reactor anomalies with an AI you can understand
The two symptoms look alike; the two responses are opposite. How a layered detection architecture turns an opaque verdict into an explainable diagnosis.
The invisible weight of a prompt: what AI really consumes
A prompt costs almost nothing. Billions of prompts cost a power plant. What can actually be done, on the hardware side, to cut the energy bill of inference.
Learning to say stop: the illusion of thinking in the age of AI
Why I decided to deliberately put friction back into the way I work, and what we lose, without noticing, when we delegate the effort before even trying to think.
MCP: when a protocol becomes infrastructure
By handing the Model Context Protocol over to a neutral foundation, Anthropic moved MCP from in-house standard to industry infrastructure. Why that changes the question we ask of an enterprise AI.