Power Platform Community Forum Thread Details

Needed your help on Copilot Studio agents that we have built in integration with Azure Devops board

(For context : we have built 3 agents : 1. Scope creep agent, 2. User story generator agent and 3. Capacity advisor agent. Connectors used are a) List iterations, b) Get query results, c) Send https request to Azure devops, d) Get workitem details, e) Create a workitem.

These connectors help us extract the needed data from Azure devops board and thereafter the agent process that data to give the required output - either scope creep, or create story in ADO, or advice on team's current capacity)

The challenges that we are facing currently are:

Issue 1: No license pooling across users (the copilot studio license that we have is not rotatable, which will make the agent person specific and once the person is out of system, agent continuity will become challenge)

Issue 2 : Scalability (For few number of stories in Azure devops, the agent works fine, but when the number of stories are high i.e. ~50 stories, then the agent output is either blank, or inaccurate

Issue 3 : How do we check the token consumption optimise tokens usage for the agent/ workflow

Issue 4 : What would be the best practices in terms of data privacy and security in the above agents context

Categories:

Calling actions from Copilot Studio

Hello SM-28061229-0,

License/continuity: Credits are tenant-pooled, but ownership & connections are user-bound—use a service account + co-owners + solution-based ALM to avoid dependency. Security and governance - Microsoft Copilot Studio | Microsoft Learn

Scalability (50+ stories): This is a payload/context limit issue. Don’t pass full query results—apply pagination, sprint/date filters, and send only key fields. Pre-process via Flow/Function and give the model a compact summary.

Token usage: You can’t see raw tokens—track via Copilot Credits in PPAC. Optimize by reducing tool calls, trimming fields, and doing deterministic processing outside the LLM. Manage Copilot Studio credits and capacity - Power Platform | Microsoft Learn

Security/privacy: Enforce DLP policies, least-privilege ADO access, OAuth/Entra ID (avoid PATs), and mask sensitive fields before sending to the agent. Custom connector frequently asked questions for Copilot Studio, Azure Logic Apps, Power Automate, and Power Apps. | Microsoft Learn

Overall pattern: Keep ADO data heavy-lifting outside Copilot Studio (Flow/Azure Function), and let the agent handle only summarized, relevant data.