Agent Loop

[TheEifelYeti]

Hi everyone,
I have a question regarding the Agent Loop in the Agent Framework.
Is there a centralized way within the framework to control or orchestrate the Agent Loop?
For example, a mechanism to centrally manage execution flow, coordination, or lifecycle of the loop across different agents?
If such a feature exists, could you point me to the relevant documentation or examples?
If not, I’d love to understand recommended patterns for implementing this kind of control.
Thanks in advance!

[Raj (rajshukla09)]

I’ve been approaching this by moving the coordination responsibility into the workflow/orchestration layer rather than having individual agents control the overall loop.
In practice, I treat agents as participants with relatively bounded responsibilities, while the workflow owns execution flow, transitions, termination conditions and coordination.

For more dynamic scenarios, I’ve also been experimenting with orchestration patterns where the next participant is selected based on the current workflow state rather than a fixed sequence.

I put together a few .NET implementations while exploring this, including dynamic and Magentic multi-agent orchestration:

https://github.com/rajshukla09/building-ai-agents-with-dotnet-part-2

One thing I’d be interested in hearing from the MAF team is whether they see this separation — agent reasoning vs workflow-level control — as the recommended production architecture as Agent Framework evolves.