Are devs really reviewing AI generated code?

Definitely. Even the current leading edge AI models (ie Claude Code Opus 4.6 (1M)) gets things (badly) wrong occasionally, makes bad decisions, and does stupid stuff.

It's a lot better than even just a few months ago, but blinding accepting its output for code that matters will catch you out.