AI Generating Full UIs: What I Delegate and What I Always Review by Hand

I asked v0 to generate a full Next.js screen and had something usable in 20 seconds — roughly what would have taken me half an hour by hand. I didn't merge it as-is.

What I do delegate to AI

• 🤖 Initial component structure and form variants
• 🤖 Repetitive loading, error, and empty states
• 🤖 A first visual draft when starting from a blank page

What I always review by hand

• 👀 Real accessibility — not just that it "looks right"
• 👀 Any logic that touches sensitive data or payments
• 👀 Whether I can actually explain why a component is built the way it is

If I can't justify a decision in the code, it doesn't ship — whether I wrote it or an agent did. AI saves me keyboard time, not judgment.