First, check Settings → Automation and confirm the master automation switch is actually on — an Agent can be assigned and enabled and still do nothing if this one switch is off.
Next, open the Audit Trail (also in Settings) and look for that Agent's most recent run. If it errored or timed out, the reason is usually right there in what it was told to do and how it responded.
If there's no recent run at all, check the thread itself — a thread already marked resolved, or one whose newest message is the Agent's own prior reply, is deliberately skipped, precisely so an Agent never loops on itself. Sending a fresh message, or reopening a resolved thread, clears both conditions.
If none of that explains it, confirm the thread isn't simply showing as already running — a second decision run is deliberately never stacked on top of one already in progress, so it may just still be working.