Agent stops responding during multi-agent orchestration workflow

(0) Share

Report

Posted on by RDAL

Hello everyone,

I'm working in copilot studio and i spotted the following issue.

My orchestrator agent becomes completely unresponsive during multi-agent workflows. It doesn't just fail to complete tasks, it stops answering to any prompts whatsoever.

Expected Behavior:
The orchestrator calls multiple agents and returns a final consolidated report.

Actual Behavior:
After a certain point in the workflow, the agent stops responding to anything. Even if I send new prompts or try to change the conversation topic, I get no response at all. The agent appears to be stuck and won't acknowledge any input.

The Core Issue:
It's not that the agent fails at its task. It's that the agent becomes completely silent, no error messages, no partial responses, nothing. It's as if the session is frozen.

Anyone ever noticed this behaviour? Thank you very much.

I apologize for the messages in italian but for your reference "genera il report" means "generate the report".

Categories:

Autonomous agents

I have the same question (0)

All responses (3)

Answers (0)

Sort by

Suggested answer

11manish 3,333 on at

Like
a
(0)

Report
Copy link

Link copied!

The orchestrator becoming completely unresponsive in a multi-agent workflow is typically caused by a deadlock situation, where one of the agent calls does not

return (due to looping, timeout, or context limits), and the orchestrator waits indefinitely.

Since no timeout or fallback is triggered, the session appears frozen with no response.

The solution is to introduce timeouts, limit agent chaining, prevent recursive calls, and add logging to identify which agent is causing the stall.

Was this reply helpful? Yes No
RDAL 2 on at

Like
a
(0)

Report
Copy link

Link copied!

@11manish Thank you very much for your answer.

Do you have suggestions for implementing these strategies without relying on prompting?

For example, can we add timeouts or logging without instruction prompts? I'm concerned the agent won't follow them reliably.

Was this reply helpful? Yes No
Suggested answer

Sayali-MSFT Microsoft Employee on at

Like
a
(0)

Report
Copy link

Link copied!

Hello RDAL,

In Copilot Studio multi-agent workflows, the orchestrator can become completely unresponsive due to orchestration deadlocks, long-running agent chains, or token/context exhaustion rather than a normal task failure. After executing multiple agent calls, the session may enter a stalled state where no further responses are generated—even for new prompts—because the conversation context becomes too large, a downstream flow does not return control, or the orchestrator is waiting indefinitely for a response. This results in a “silent freeze” with no errors or outputs. The issue is commonly observed in complex multi-agent chains, especially when flows are repeated, responses are large, or no timeout/fallback handling is implemented.

The recommended approach is to simplify orchestration, limit agent chaining depth, introduce explicit timeouts and fallback responses, reduce payload size (e.g., large JSON outputs), and ensure each agent returns a clean, bounded response to the orchestrator.

Was this reply helpful? Yes No