When I work with AI tools for building products, the hard part is never "can AI write this function" - it almost always can. The hard part is knowing what to build, recognizing when the generated code has subtle bugs, understanding the second-order effects of architectural decisions, and having the pattern recognition to say "this will become a maintenance nightmare in 6 months."
I think "vibe coding" exposes who was always just a syntax memorizer versus who actually understood systems. The former group will struggle because their value was in translating requirements into code - AI does that now. The latter group will be more productive because they can skip the tedious parts and focus on design, debugging, and judgment calls.
The analogy I keep coming back to: calculators did not replace mathematicians. They replaced people who could only do arithmetic.