Make MAF agents reliable with Dapr Workflow

Agents call large language models, and LLM calls are slow, costly, and non-deterministic. When a multi-agent app crashes halfway through, re-running everything from scratch wastes time and money. In this self-paced track you'll see how Dapr Workflow turns a collection of Microsoft Agent Framework (MAF) agents into a durable, fault-tolerant application.

What you'll run

You'll work with PrDigest, a .NET Aspire app that triages open pull requests for the Dapr project: a PrAnalyzer agent analyzes each PR, the workflow ranks them by risk, and a Summarize agent writes a headline digest. You'll trigger the workflow, crash it mid-run, and watch it resume from durable state — replaying the agent calls that already completed instead of calling the LLM again. You'll need around 25 minutes to complete the 3 challenges.

Ensure you have your own OpenAI API key before you start this track.