Power Platform Community Forum Thread Details

Hi everyone,

I'm currently building a unified agent experience in Copilot Studio, where I’ve connected several standalone agents (now acting as sub-agents) under a main agent. The routing works well—user queries are correctly directed to the appropriate sub-agent based on intent.

However, I’m facing a problem where, within the sub-agent, when it tries to perform a database search via a connector, the process gets interrupted and I see the message "Picked by GPT" instead of the expected connector response.

What I’ve Observed

The sub-agent was originally built as a standalone agent and worked fine on its own.
After connecting it to the main agent, it still routes correctly and starts processing.
But when it reaches the connector step (e.g., SQL or SharePoint), it gets overridden by GPT and doesn’t complete the search.
This behavior happens within the sub-agent, not during the handoff from the main agent.

What I Want to Achieve

I want to keep orchestration enabled in the sub-agent (so it can intelligently decide how to respond).
But I want to avoid infinite loops or GPT overriding the connector logic.
Ideally, once the sub-agent completes its task, it should return a final answer back to the main agent cleanly.

What I’ve Tried

Ensured the sub-agent has a clear topic flow and connector setup.
Verified that the connector works when tested independently.
Checked orchestration settings but still seeing GPT take over.

Questions

How can I prevent GPT from interrupting the sub-agent’s connector actions while keeping orchestration active?
Is there a way to force the sub-agent to return a final answer to the main agent without looping or fallback?
Are there best practices for designing sub-agent exit flows in a unified agent setup?

Any insights, tips, or examples would be greatly appreciated!

Thanks in advance 🙏

Categories:

General topics