As AI-driven automation rapidly evolves, developers face a critical challenge: how to orchestrate and manage intelligent agents at scale in a way that’s secure, reliable, and production-ready. In our latest Open at Microsoft episode, we dive into an exciting new open-source framework that’s designed to tackle exactly that—Dapr Durable AI Agent Workflow Framework, or simply, Dapr Agents.
Yaron Schneider, CTO and Co-founder of Diagrid and original co-creator of Dapr, walks us through why Dapr Agents is a game-changer for building agentic systems and how it integrates seamlessly with the CNCF Dapr runtime.
What is Dapr Agents?
Dapr Agents is an open-source framework purpose-built to simplify the development of durable, stateful, and production-grade AI agents. It extends the power of Dapr—a CNCF-graduated distributed application runtime used in mission-critical workloads worldwide—by providing the enterprise capabilities developers need to deploy autonomous multi-agent systems with confidence.
Rather than reinvent the wheel, Dapr Agents leverages Dapr’s existing building blocks: virtual actors, pub/sub messaging, state stores, workflows, service discovery, and more. This makes it incredibly easy for developers to go from a local prototype to production-ready agentic infrastructure without needing deep expertise in distributed systems.
Why Agentic Systems Need More Than Just LLMs
Most agent frameworks today are great at letting you run a single LLM-powered agent or a simple task orchestration. But as Yaron points out, the moment you try to scale that to a team of agents, run in a distributed environment, or plug it into enterprise systems—things get complicated fast.
Key challenges include:
- Durability: If an agent fails at step 99 of 100, how does it resume without restarting the entire process?
- Security: How do you ensure authenticated, authorized agent-to-agent and agent-to-service communication?
- Observability: Can you trace and audit what each agent did and when?
- State Management: Can agents remember what happened previously and make decisions based on shared or personal state?
- Collaboration: How do agents coordinate across workflows and communicate reliably?
Dapr Agents addresses all of this out of the box.
Frodo, Legolas, and Gandalf Walk into a Workflow…
One of the highlights of the episode is a delightful and educational live demo. Yaron walks us through a multi-agent system modeled after the Fellowship of the Ring. Using Dapr Agents, he spins up an “agent mesh” made up of Frodo, Legolas, and Gandalf—each with specific roles and instructions—to collaboratively build a plan to take the One Ring to Mordor.
The key takeaway? Developers only need to define roles, goals, and simple instructions. Everything else—messaging, retries, state, durability, and security—is handled by Dapr Agents. And it’s all pluggable into the infrastructure you’re already using—whether that’s Azure Cosmos DB, Redis, Kafka, or your preferred tracing system.
Enterprise-Grade Capabilities Built-In
Unlike many agent frameworks that stop at orchestrating LLMs, Dapr Agents was designed with real-world production deployments in mind. Here’s what it brings to the table:
- 🔐 Authentication & Authorization – Built-in identity propagation across agent calls and external tools (like MCP), with OAuth2 and enterprise SSO support.
- 🧠 Durable Workflows – Agents pick up where they left off—even after failure.
- 📡 Event-Driven Agents – Ambient agents can be triggered by real-time events, not just prompts.
- 📊 Observability – Native support for tracing (Zipkin, Azure Monitor, etc.) without adding code.
- 🌐 Vendor-Neutral – Co-maintained by Diagrid and NVIDIA, it’s the only CNCF agent framework of its kind.
Get Involved
Dapr Agents is open-source and growing fast—and the community plays a vital role in shaping its future. If you’re passionate about building agentic systems or want to get involved:
👉 Star the repo: github.com/dapr-sandbox/dapr-agents
💬 Join the conversation on the Dapr Discord
🛠️ Open an issue or contribute code and ideas to the roadmap
📚 Check out the docs and get started with the quickstart demos
As the world pushes deeper into autonomous systems powered by LLMs, frameworks like Dapr Agents will be essential in bridging the gap between proof-of-concept and production. With Dapr, developers don’t have to build observability, resilience, and state management from scratch—it’s all there, ready to support intelligent, scalable agentic systems.
📺 Watch the episode now: Open at Microsoft – Dapr Durable AI Agent Workflow Framework
📦 Get the code and contribute: Dapr Agents GitHub
