I have now been using the "Generative Mode (Preview)" (i.e., not the Classic Mode) for some months to create a small agent with about 5 actions, 10 additional topics, and 5 knowledge sources (uploaded documents). My experience is that this Generative Mode does not work well at all (in contrast to the demos at the Ignite event in last November). Most of the time, the agent does simply not recognize that it has a suitable action or topic even if I input exactly the sentences which the actions and topics are made for. It seems to me that the agent often just searches in the knowledge sources without considering actions or topics. And when I am asking a question which exactly matches the description of a knowledge source then often the agent does not recognize that there is a suitable knowledge source to answer the questions. And I experience this despite the fact that I have invested lots of time to try many, many difference system prompts and action/topic/knowledge source descriptions without overlapping scope.
And yes... I know that this Generative Mode ist just preview... but nevertheless I am a little bit surprise about the low quality because a) it has been in preview for soooo long so I would have expected this to be soon ready for production and b) I have seen the message "From February 1st, preview features will be billable and consume messages." in my Copilot Studio which should mean that the preview feature are so good that users have now to pay.
Thing is that I see that other no code agent platforms manage these things way better and I am asking myself whether I should still wait until Microsoft gets the Generative Mode right or whether it's time to switch to another platform.
What's your view? What is your experience with the Generative Mode? Do you experience similar problems or is all working smoothly for you?
Thank you for your reply! To ensure a great experience for everyone, your content is awaiting approval by our Community Managers. Please check back later.