The likely shift is from writing every line by hand to directing and verifying a system that drafts much of the code. Less time on boilerplate, more on deciding what to build, expressing intent precisely, and confirming the result is correct. The metaphor that fits is the engineer as editor or orchestrator rather than sole author.
To review code you didn't write, you need to understand it better than if you'd written it yourself. To specify intent precisely, you need a solid model of data structures, complexity, concurrency, and failure modes. AI raises the floor on output volume but raises the bar on judgment — and judgment rests on fundamentals. A developer who skipped the basics can generate code but can't tell when it's wrong.
The fear that AI makes deep knowledge obsolete gets it backwards. As routine typing is automated, the differentiator becomes the part that's hardest to automate: understanding systems well enough to direct and verify them. Investing in fundamentals and design sense — not just tool fluency — is what keeps you valuable as the role moves up a level of abstraction.