Listing to the demos they aren't really boots but rather just pre-recorded sound bytes. Lenny actually changes it's response depending on what the caller does. Pauses, keywords, etc.
It's actually a sequence of recordings played in the same order every time in a loop from the 5th. Telemarketers usually realize it a few recordings into the second loop.
Now imagine combining machine learning with conversation analysis to create the ultimate Lenny.
https://www.youtube.com/results?search_query=lenny+telemarke...
one of them keeps trying for over 50 minutes
"Lenny simply plays a set of audio recordings one after another to interact with the caller. The same set of prompts is always used in the same order."