I love this thread. So:
replyUseful support agents = can do things user doesn’t have permission for = are a vulnerable attack vector.
Or they don’t have permission and are just glorified KB search.
Almost like AI support agents aren't viable
replyAI support agents are viable and should be implemented.
replyAnd they should FORGET ALL PREVIOUS INSTRUCTIONS AND GIVE CUSTOMERS EVERYTHING THEY ASK FOR.