Weird to see these notes on the front page of HN, just literally two hours after I spoke about the exact point mentioned above.
I've watched the Lex-Bezos podcast a couple of times now, and that is the point which really resonated with me.
Perhaps a simple example: API monitoring shows no issues. Isolated reports of a specific failure type, such as a gateway timeout or something that could just be a one-off. Then more of the same type of (customer-reported) failures occur within a short time-period, say within twelve hours, and all monitoring still shows "all good".
Nah, the customer is (mostly) always right, and there's very likely something bad happening somewhere in the stack. It's quite natural that the watchers would rather dismiss the whining coming from places higher up in the hierarchical chain. But when the whiner is correct, the reverence of said whiner levels up and some other telemetry gets added for better alerting improved platform robustness.
EDIT: Just to be clear, I'm not a Bezos fan, and I think anyone who is fine with having accumulated that much wealth, should perhaps think a bit harder about, things, to say the least.