Best Practices for Combining Temporal with Autogen for AI-Driven Workflows

Hi everyone,

I’m exploring a hybrid approach that integrates Temporal with Autogen Core (v0.4) to orchestrate AI-powered workflows. The idea is to use Temporal for fault-tolerant execution while leveraging Autogen as a multi-agent communication framework.

Proposed Architecture:

Temporal Workflow: Manages the orchestration of AI tasks, ensuring state persistence, retries, and human-in-the-loop approvals.
Autogen Agents as Temporal Activities: AI logic (e.g., reasoning, collaboration, and API calls) is handled within Temporal activities using Autogen.Runtime to manage multiple agents.

Key Questions:

  1. Is this a valid and recommended approach?
  2. How should I handle Autogen.Runtime? Since Temporal workflows need to be deterministic, would it be better to reinitialize the runtime within each activity rather than passing it across calls?
  3. Are there potential pitfalls to watch out for?

Looking forward to insights from the community!

Thanks!