Your human in the loop needs a paper trail
Research described in MIT Technology Review found that the four most widely used agent frameworks don't reliably record a persistent agent ID or which human approved a task. The second gap is the one that worries me. When I helped operationalize a GDPR program, the first question in every review was who decided and on what basis. If your framework can't show which officer signed off on an agent's recommendation, your human in the loop is a story you tell auditors. The authors are right that banks and hospitals can push vendors on this. Write those records into procurement requirements before any pilot reaches production. Could your agents survive that audit?
Source: We urgently need to monitor AI agents’ actions in real time (MIT Technology Review)