The PC Pro Podcast Episode 782: Sure, the code took Amazon down for six hours, but at least it’s commented
Mar 12, 2026
They dig into Amazon’s six-hour AI coding outage and the new human-approval rule. They debate whether AI-generated comments can hide risky code and recount AI-created fake references on Wikipedia. The idea of a unified UK streaming iPlayer and Norway’s manifesto to fight tech decline are discussed. They also spotlight an AI live-transcription tool that turns speech into polished prose.
AI Snips
Chapters
Transcript
Episode notes
AI Code Needs Senior Human Approval
- AI-assisted code can introduce subtle, production‑level risks even when it appears well documented.
- Amazon experienced a six-hour retail outage and now requires senior human approval for any AI-generated code changes to prevent visible failures.
Coder Compares AI Code Risk To Asbestos
- A PC Pro reader likened AI-generated code risk to asbestos, warning damage might take decades to surface.
- The reader and colleagues find themselves auditing AI output for a living, suggesting hidden long-term liabilities in software stacks.
AI Translations Can Fabricate References
- AI-assisted translation can silently introduce fabricated references and facts into published content.
- Wikipedia saw translated articles with invented citations after translators were encouraged to use AI tools without strict defensive prompting.