AI hasn't "forced" domain knowledge to be required at all, it's been there as a requirement for a long time (I had several medical projects turn myself and many other good developers away because we didn't have domain knowledge, which was fair because we needed to have some idea of the subtext in the field)
The famous essay "Don't Call Yourself A Programmer" dates from 2011! I thought it was required reading here on HN.
https://www.kalzumeus.com/2011/10/28/dont-call-yourself-a-pr...
Anyone seriously interfacing with AI code generation can tell you that understanding how software is composed and architected is, and will continue to be, a requirement to building anything worthwhile. These systems are impressive but we are a far ways away from "Magical Genie Wishes" as a means of creating software.
Do they spring fully formed from Zeus' forehead?
In fact, AI code generation hasn't changed much at all for anyone with a significant amount of experience (outside of business and management expectations) - its just another tool of varying usefulness.
After my recent layoff, I realized I had become too narrow. I knew one language and one subject matter well, and couldn't quickly switch to something else. I was lucky and found an exact fit within a week - but there's no guarantee that will happen again.
Definitely doesn't hurt to know a few languages, a few subjects well. Even if you're not an expert, you can fake it for a bit while you get the details down.
pjmlp•1h ago