The interesting part is not that it built a driver, it is that it built one for a specific chip on a specific OS with no existing example to copy. That is not autocomplete. That is reasoning about hardware registers, kernel interfaces, and driver architecture from documentation and first principles. A year ago this would have been a strong counterexample to AI coding claims. Now it is a blog post.

What? No. An LLM cannot reason, at least not what we think of when we say a human can reason. (There are models called "reasoning" models as a marketing gimmick.)

TFA describes a port of a Linux driver that was literally "an existing example to copy".