Power Platform Community Forum Thread Details

We are currently taking our first steps in developing and deploying custom agents using Microsoft Copilot Studio, and I would like to clarify the licensing and access model for end users.

Our scenario:

A limited number of users build and manage custom agents in Copilot Studio.
The agent is published within our corporate Microsoft environment.
The agent uses Dataverse tables and Power Automate flows to retrieve and process information.
A larger group of end users will only interact with the published agent; they will not create or manage agents.
These end users have our standard Microsoft 365 setup and access to Copilot Chat, but do not necessarily have an individual Microsoft 365 Copilot or Copilot Studio license.

My questions are:

Do these end users need an additional Copilot Studio or Microsoft 365 Copilot license to fully use the published custom agent?
If not, can their usage be centrally covered through the Copilot Credits/capacity or pay-as-you-go model configured for the Copilot Studio/Power Platform environment?
If two users access exactly the same published agent, ask the same question, and have identical permissions to the underlying Dataverse data and Power Automate flows, should the agent provide the same functionality and comparable response quality regardless of whether one user has a Microsoft 365 Copilot license and the other only has standard Copilot Chat access?
What are the minimum permissions and/or licenses an end user needs for the Power Platform environment, Dataverse tables, connections and Power Automate flows used by the agent?
Is the following deployment model therefore supported?

A limited number of licensed Copilot Studio makers create and manage the agents, while a much larger group of standard Microsoft 365 users consume the published agents, with agent consumption being centrally billed through Copilot Studio Credits/capacity or pay-as-you-go.

The reason for my question is that during testing we have observed significantly different results between users accessing the same published agent. One user receives a detailed, properly grounded answer, while another user receives a much shorter and less useful response that appears not to use the underlying data sources.

I am therefore trying to determine whether this difference can be caused by the end user's license, or whether we should focus entirely on environment access, authentication, Dataverse security roles, connection permissions, or other configuration settings.

Any clarification or experience with this type of enterprise deployment would be greatly appreciated.

Categories:

Autonomous agents

Hi,

Yes, the deployment model you described is supported. A small group of licensed makers can build and manage agents, while a larger group of users consumes those agents through centrally managed Copilot Studio capacity or pay-as-you-go billing.
For your specific questions:

End users do not necessarily need a Copilot Studio license just to use a published agent, but they do need access to the channel where the agent is published and the underlying resources it uses.
Usage can be covered through Copilot Studio capacity/credits or pay-as-you-go, depending on how your environment is configured.
If two users have identical permissions and access to the same data sources, they should generally receive comparable results. Significant differences are more often related to permissions, authentication, Dataverse security roles, connection references, knowledge source access, or grounding configuration than to whether one user has a Microsoft 365 Copilot license.
Users must have the appropriate permissions to the Power Platform environment, Dataverse tables, and any resources being accessed through flows or connectors. Missing permissions can cause the agent to return reduced, incomplete, or ungrounded responses.

Given the behavior you're seeing, I'd focus first on security and access configuration rather than licensing. Verify that both users have the same Dataverse security roles, access to the underlying SharePoint/knowledge sources, flow permissions, and connector authentication context. Those differences are commonly responsible for inconsistent responses from the same published agent.

Thanks & Regards,

Manoj Annavajjala