Skip the feature grid and ask four things: does the agent act, or only suggest? Does it work in your channels, or only in-app? Is it shaped to your motion, or templated? And when it's unsure, does it escalate cleanly?
A CRM with a chat box can fail all four and still demo well. An agentic system has to pass them, because passing them is what it is. The checklist is short on purpose.