The AI Agent waits a short interval before replying. In exceptional cases, if the contact sends a second message right when the first reply is already being delivered, two consecutive messages can arrive.
How the AI Agent processes messages
Before generating a reply, the agent waits between 10 and 15 seconds since the contact’s last message. This lets it group several short messages sent in a row and reply with a single, coherent message.
This mechanism is called debounce. It makes the agent sound natural instead of replying message by message.
When two replies happen
This happens when the contact sends a second message after the debounce window (~15 seconds) has already passed:
- The agent receives the first message and, after the wait, starts generating the reply.
- While the reply is being delivered, the contact sends a second message.
- At that point the first reply can no longer be canceled — it’s already in delivery.
- The agent processes the second message and generates a second reply.
Real example: a contact writes “I want information about the service” and 22 seconds later writes “WhatsApp only”. The agent replies to the first one and, on detecting the second one outside the window, also generates a second, adapted reply.
Is this a bug?
No. It’s expected behavior in a very specific edge case. It happens when the second message arrives at the exact moment the first reply is already being delivered and can no longer be canceled.
It’s rare and only occurs when both conditions are met at once:
- The contact takes more than ~15 seconds between messages
- The second message arrives right while the first reply is being delivered
Why isn’t this fixed by increasing the wait time?
Increasing the debounce window would reduce this case, but it would make the agent slower to reply in every other case too, including normal ones. The product team has prioritized reply speed in the general case over eliminating this rare exception.
Conclusion
The two replies don’t indicate a failure of the agent — they’re consistent with each other and respond to real messages from the contact. If a customer reports it, you can explain that it’s a one-off behavior with no impact on conversation quality.
Relevant scenarios
- Contact writing in two separate moments: if there are more than 15 seconds between messages, it’s normal for them to receive two independent replies.
- Customer reporting a double message as a bug: it’s not an error — the agent replied to two different messages that arrived outside the grouping window.
- Agent appearing confused or repetitive: the two replies are independent, coherent processes, there’s no confusion in processing.
Was this article helpful?