I built a few AI Agents, and then I built a Master Agent that delegates requests to those other agents.
The advantage is that I only need to deploy the Master Agent, and as new capabilities are added, the Master Agent will simply get updated and offer new capabilities.
(I am deploying the Master Agent to Teams.)
If I say "Prospect for XXX", that will run the Prospect Agent giving the XXX as parameters. The Prospect Agent is rerunnable -- you can run it over and over again. The data in the underlying systems may change, so you would get the latest results each time.
My problem is very simple: when I ask the Master Agent to run "Prospect for XXX" a 2nd (or nth time), it simply responds with a message like, "Prospecting has been completed..." In other words, there is a safeguard somewhere that is trying to avoid repetitive requests. Alas, I do not want that -- I need to be able to run this command more than once.
A common scenario is that the first request essentially canceled because it was missing a data element. So I go and add that data element, and then want to run the request again, but the second attempt is rejected.
I have tried to modify the Master Agent instructions to emphasize that I want to re-run requests. However, I'm coming to the conclusion that Copilot Studio must have some hidden prompt that is preventing the duplicate request.
Is there a workaround? How can Copilot Studio get fixed?