When AI Writes 95% of Your Code, Developers Still Aren't Obsolete

Just read about Coinbase's 95% AI-generated code. First thought: "How do they even review this?" Because without senior engineers who can spot garbage code, you could generate 150% AI code and still ship trash.

AI Doesn't Write Code. It Suggests Code

I'm literally using Cursor right now to refactor legacy code. Yes, AI saves me hours—but only because:

1. I know what the end result should look like
3. I can instantly catch weird suggestions like if (value === undefined || value === null) instead of proper if (value == null)

4. Without these? You've got a fancy random code generator.

   Where AI Actually Shines

   Here's what AI genuinely rocks at:

   • Polishing boilerplate (React's useEffect, anyone?)
   • Generating test mocks
   • Quick edge case spotting ("what if the array's empty?")

   • But architecture, complex logic, and especially code reviews? Still human territory. Or does Coinbase now have AI reviewing AI? I'd kill to see their internal processes.

     P.S. My personal record: 3 hours for a refactor that'd take a day without AI. But only because I knew EXACTLY how to do it right.