Your model can produce French. That does not mean it learned to reason with French assumptions. This piece looks at English-dominant training, multilingual reasoning, and the question still missing from most sovereign AI debates.
On language, cognition, and what we are encoding when we build AI in English.
China's courts ruled that firing workers to replace them with AI is illegal. The EU spent months negotiating and landed on a ban on nudification. The US is preempting the states from trying to do anything at all.
AI sovereignty does not stop at infrastructure. It shows up in product architecture: where inference runs, how data flows, whether models can be replaced, and how much of your product depends on someone else’s roadmap.
Your model can produce French. That does not mean it learned to reason with French assumptions. This piece looks at English-dominant training, multilingual reasoning, and the question still missing from most sovereign AI debates.
On language, cognition, and what we are encoding when we build AI in English.
China's courts ruled that firing workers to replace them with AI is illegal. The EU spent months negotiating and landed on a ban on nudification. The US is preempting the states from trying to do anything at all.
The AI conversation assumes everyone wants to become a builder. Most people do not. And the workers most exposed to disruption are the ones least represented in the rooms where AI strategy gets decided.
I was recently interviewed by Roberto V. Zicari for ODBMS Industry Watch, a publication I follow and respect for the seriousness of its conversations. The full interview is linked at the bottom. But a few things I said there are worth pulling out here, because they are things I keep
AI sovereignty does not stop at infrastructure. It shows up in product architecture: where inference runs, how data flows, whether models can be replaced, and how much of your product depends on someone else’s roadmap.
AI sovereignty is not a binary or a product. It is a question of control, dependency, and what organizations are willing to give away by default.
A reflection on what AI changes when building and shipping products becomes dramatically easier, and why speed does not replace product judgment.
A few years after this interview, I still find myself coming back to the same questions: representation, leadership, defaults, and who gets to shape the systems we build.
Why multilingual AI still too often means English-first AI, and why that matters for access, quality, and who gets to build.