AI Email Messages

[deycom]

Hi there,

I've been using the new AI feature to send a message by email to the email address associated with an extension number in our 3CX system. The AI, was taking the message, and then at the end of the call confirming the message had been sent. However after a recent update, the call is terminated immediately after the email is sent. The email messages are still being sent and received.

I'm not sure if the email function is still in testing?

Any help is much appreciated.

Thanks,

Bryan

 

[accentlogic]

@deycom I noticed this today. I believe email action option is missing the confirmation described in the AI SUMMARY Suggested Steps sent via email. After the action the next step is end call.

- Suggested Steps
I stated I would send an email to %Firstname% %Lastname% and let %Firstname% know.

@nikolascx if this is correct, I believe the AI Agent should play the statement back to the caller and then politely announce the call is ending or ask them if they would like any additional assistance.

 

[nikolascx]

Hi Bryan,

Yes, your understanding is correct. And to be transparent, this is something we are aware of as well. It is a legitimate temporary trade-off, and I’ll explain why.

At the moment, after the AI Agent sends the email, the call is intentionally ended. The email action should be treated as completed once it has been triggered, so the messages will still be sent.

We made this change because the post-email confirmation step was not always predictable. In some cases, asking the AI Agent to perform another action after sending the email, such as confirming “your email has been sent, goodbye”, could interfere with the end-call flow and leave the call connected.

When that happened, the AI could resume unnecessary conversation with phrases like “Are you still there?” or “Can I help you with anything else?” This kept calls open longer than needed, wasted tokens, and effectively reopened a conversation that should already have been closed.

So for now, this was a deliberate trade-off. We sacrificed the polite post-action confirmation in order to make the email action and call termination more predictable and reliable, and to avoid these kinds of phantom calls.

We absolutely understand the value of confirming the action back to the caller, or asking whether they need anything else, and we plan to revisit these courteous post-action confirmations later. This is temporary. Tool calling in AI models is improving quickly with each new model update, and as the behavior becomes more reliable, we’ll be able to improve this flow further.

Thanks again for flagging it and for the useful feedback.

 

[deycom]

Thank you for the feedback & suggestions - will keep an eye out for future updates and developments