Perspectives from AI practitioners on risk, unexpected behaviors, and what it takes to trust agents in production
