Amazon now requires engineers to get a two-person review before making any changes to production, and junior engineers must get AI-generated code reviewed by senior engineers.

Amazon experienced four critical errors in just 90 days due to AI-assisted code changes.

Recent studies have shown that AI-generated code contains up to 1.7 times more errors than code written by humans.

Soleyman believes Amazon is the best example of how things can go horribly wrong when replacing engineers with AI that is not ready.

Amazon experienced 13 hours of downtime for AWS when an engineer used a mandated AI coding tool that deleted the entire production environment.

Amazon's assessment system logs browser activity, restricts copy-paste, and allows public resources but not anything behind a login, which rules out most AI tools.