I still code by hand the majority of the time. Not because it is faster, or more productive, but because it is the medium to learning and truly understanding something.
Let's say I want to learn how data is sent over the web (and when I say learn, i mean truly understand, at a deep level how something works). To do so, I should probably build an HTTP server from scratch.
I might choose a language like Go, start coding HTTP streams, parsing request lines, dealing with chunk encoding etc. I would have to work out on my own how each component of an HTTP server works. This is an active process and where the learning happens. Eventually, upon a completed HTTP server I would know how the building blocks stack upon one another to send data across the web.
The point is this is an active process. Hands on keyboard, working through the problem space until it is solved. The goal isn't to produce something meaningful, or quickly. But it's to learn and deepen my understading of a fundamental technology of the web.
My job is to be knowledgable on tech and it's what people in the industry get paid for. If I had AI spin up an HTTP server for me instead, a whole level of understanding would be lost.
Therefore, unless I know something well enough already and I don't just need something working for me as quickly as possible, I will code by hand and leave the clanker to the side.
The distinction here is between professional output and personal knowledge / growth. For the former, I find AI a perfectly acceptable tool. For the latter, I don't. I think too many people are conflating the two worlds now.