What’s harder today: writing code or reviewing AI-generated code?

I find myself spending less time creating code from scratch and more time validating assumptions, edge cases, and architecture decisions.

The workflow feels very different from even a year ago.