AI writes the first draft now. I built practice for the part that's left: pushing back on it

Hi PH, I'm building Prepd, an interview practice site for software engineers.

Something I keep seeing: producing code and architecture diagrams is getting cheap, because an assistant gives you a decent first draft in seconds. What's getting more valuable is the review. Spotting the off-by-one, the shared state with no lock, the API that doesn't exist, or the design with Kafka and five services for 50 writes a second, and then saying "this is wrong, here's why" clearly.

AI-assisted interview rounds test exactly that. So Prepd now has: AI-written code with planted bugs to find and fix, system design drills with written feedback on your trade-offs, and a free animated learning path that shows what each scaling step fixes and what it breaks.

I also wrote up a free checklist for reviewing AI-generated code and designs, with phrases for pushing back and two worked examples: https://prepd.tech/guides/reviewing-ai-generated-code-and-designs

Curious how others handle this: when an AI suggestion looks right, what's the first thing you check before accepting it?