We built an elaborate apparatus for turning AI models on safely - reviews, red teams, evals, benchmarks. To turn one off took a letter that landed at 5:21pm on a Friday. The model was never the thing that needed governing. The switch was.

The détente is over. Anyone can write software now—and soon, anyone can break it. What happens when the people building the internet are the least equipped to defend it?

My Tesla crashed with my kids in the back seat. Here's what I've been thinking about since.

One more from the India AI Summit - on why sovereignty doesn't mean solitude, and why the countries that wait to invest in open AI are the ones that won't get to choose later.

On my way to the India AI Summit with a simple argument: when open-source AI matches proprietary performance at a sixth of the cost, sovereignty stops being ideological and starts being practical.

Humanity thrives on friction—so why are the tools of the future built to make everything seem so easy?

On raising the first AI-native generation, and the human skills—resilience, questioning, empathy—that will matter most.

A call to humanize government technology—proving that efficiency and empathy can, and must, coexist.

A look at how self-driving cars are turning our cities into networks of private surveillance—and why that should concern us all.

A proposal for an Internet of Things and its impact on infrastructure and everyday services.