You did something in an app - submitted an invoice, advanced a deal stage, filed an expense claim - and you expected the AI agent to update the record. It didn't. Here's what to check, in order.
1. Did the agent actually fire?
On the record itself, look for the Activity tab (or Audit / History tab - the exact label varies by app). If the agent fired, there's a row that says "Agent run started: <agent name>" - click it to see what happened.
2. Did it pause for human review?
If the activity row says "paused for review" instead of "completed", a HITL guardrail tripped. The run is waiting for a human approval - check Sidebar > HITL Reviews. If you're the assigned reviewer, you'll see a pending one to approve/reject.
3. Did it fail?
If the activity row says "failed", click it. The error message explains why - common: a tool returned an error, the agent's confidence was below threshold, an external system was unreachable.
4. Did it actually finish, but you missed the update?
Hard refresh the page (Cmd/Ctrl + Shift + R) - some agent outputs are written to fields you might not be looking at. The Activity tab is the source of truth.
5. Is the agent currently disabled?
Org Admins can pause an agent (Sidebar > Agents > [agent] > Pause). If it's paused, no runs fire. Ask your admin if you suspect this.