The artificial intelligence (AI) agent landscape is rapidly evolving, but a critical gap remains between demonstration-worthy prototypes and production-ready systems. Yaron Schneider, Co-Creator of Dapr, believes his team has found the solution with Dapr Agents — a framework that brings enterprise-grade durability to AI-enabled autonomous agents running on Kubernetes.

The Dapr Foundation: Simplifying Application Development

Before diving into Dapr Agents, it’s essential to understand what makes Dapr unique within the Cloud Native Computing Foundation (CNCF) ecosystem. Unlike most CNCF projects that target infrastructure professionals, Dapr specifically addresses application developers’ pain points.

“Dapr is a unique project within the CNCF in that it really speaks to application developers, as opposed to infrastructure people,” Schneider explains. “When developers need to write applications on top of Kubernetes, there is a lot going on — they need to understand how to talk to different databases, message brokers, secret stores, caches, how to write long-running, stateful workflows — and that is extremely complicated.”

Dapr abstracts this complexity by providing developers with a set of APIs that encapsulate the hard boilerplate code and logic, allowing them to focus on their intellectual property rather than reinventing infrastructure wheels.

The Durability Problem in AI Agent Frameworks

The launch of Dapr Agents addresses a fundamental limitation in current AI agent frameworks. While existing solutions excel at proof-of-concepts and demonstrations, they struggle with the unpredictable nature of AI agents in production environments.

“AI agents really become useful when you give them full autonomy,” Schneider notes. “You don’t want to have to maintain an agent and make sure that it actually completes its task. But AI agents are also unpredictable.”

The core challenge lies in ensuring reliable execution when granting increased autonomy to AI systems. Current frameworks lack the resilient building blocks necessary for production deployment, particularly when dealing with infrastructure failures, network interruptions, or scaling events.

Pain Points Driving Innovation

The creation of Dapr Agents stemmed from real-world developer experiences with existing agentic frameworks. Schneider describes a pattern where developers would successfully run proof-of-concepts but encounter significant issues when moving to pre-production or staging environments.

“As the traffic increased, or, you know, pods went down, or they scaled their cluster, they would see that the agents actually got disrupted and were unable to complete their execution,” he explains.

This reliability gap prompted a Microsoft contributor to develop the first version of Dapr Agents, which was subsequently contributed to the CNCF and announced by the Dapr project approximately a month ago.

Technical Advantages: Durability Meets AI

Dapr Agents distinguishes itself by building upon the already battle-tested Dapr workflows framework. This foundation provides crucial durability features that other AI agent frameworks lack.

“Imagine you’re running a complex task for an agent, and your cluster shuts down, or you have a network interruption, or your Kubernetes pod goes down — Dapr Agents will actually spin back up, it will rehydrate its state, and it will continue where it left off,” Schneider explains.

This capability combines AI functionality with the resiliency and durability expected from deterministic applications, making autonomous AI agents truly production-ready.

Real-World Applications Emerging

Despite being relatively new, Dapr Agents is already seeing adoption across diverse use cases. Early implementations span from DevOps automation — including agents that automate Kubernetes clusters — to customer success representatives that connect to enterprise data sources and solve complex user problems.

Infrastructure automation represents another significant application area, with agents capable of automating CI/CD pipelines. The framework’s general-purpose nature enables this diverse range of implementations, reflecting the broad applicability of resilient AI agents across different business functions.

Community Response and Future Roadmap

The Dapr community is actively providing feedback that’s shaping the framework’s development roadmap. A key request emerging from users is integration with Anthropic’s Model Context Protocol (MCP), which would bring enterprise data directly into agentic systems for improved decision-making.

“The CNCF ecosystem is currently, in my opinion, very far removed from the AI ecosystem. And so what we want to do is really bring those two together,” Schneider states.

The MCP integration represents a crucial first step in bridging this gap and bringing AI workloads into CNCF projects like Dapr.

Language support expansion is another priority, with plans to extend beyond Python to include C# and Java. This expansion acknowledges that while Python dominates machine learning development, application developers building production agents often come from line-of-business backgrounds where C# and Java are more prevalent.

Bridging Cloud-Native and AI Ecosystems

Dapr Agents represents more than just another AI framework — it’s a bridge between the established cloud-native ecosystem and the emerging AI landscape. By providing the durability and resilience that production AI agents require, while maintaining the developer-friendly approach that made Dapr successful, this framework could accelerate the adoption of autonomous AI systems in enterprise environments.

As organizations increasingly seek to deploy AI agents in production scenarios, the combination of Kubernetes orchestration, Dapr’s application runtime capabilities, and AI agent frameworks creates a compelling foundation for the next generation of autonomous systems.

The success of Dapr Agents may well determine how quickly the promise of autonomous AI agents transitions from compelling demonstrations to reliable, production-grade solutions that enterprises can depend upon.

Edited Transcript

Swapnil Bhartiya: Hi. This is Swapnil Bhartiya, and welcome to KubeCon + CloudNativeCon in London. Today, we have with us Yaron Schneider, Co-Creator of Dapr. Yaron, it’s great to have you on the show.

Yaron Schneider: Thank you so much for having me. It’s my pleasure to be here.

Swapnil Bhartiya: You are, of course, Co-Creator of Dapr. Can you talk a bit about what Dapr is all about?

Yaron Schneider: Dapr is a unique project within the CNCF in that it really speaks to application developers, as opposed to infrastructure people. When developers need to write applications on top of Kubernetes, there is a lot going on—they need to understand how to talk to different databases, message brokers, secret stores, caches, how to write long-running, stateful workflows—and that is extremely complicated. Dapr gives developers a set of APIs they can use that encapsulate all of this boilerplate code and logic, so developers can focus on their own IP and what matters to them instead of needing to reinvent the wheel time and time again.

Swapnil Bhartiya: Today we’re also going to talk about Dapr Agents. Can you tell us what Dapr Agents is?

Yaron Schneider: Dapr Agents is a framework for developers to build AI-enabled agents that run autonomously on top of Kubernetes and outside of Kubernetes. You hear a lot of talk about other AI agent frameworks out there today. What makes Dapr Agents really special is that it builds on top of the already battle-tested and proven Dapr workflows.

AI agents really become useful when you give them full autonomy—you don’t want to maintain an agent manually or worry about whether it completes its task. But AI agents are also unpredictable. If we’re giving more responsibility to an AI agent, how do we ensure it drives its scenario to full, successful execution?

Many AI agent frameworks today don’t have the resiliency required for that. They’re great for POCs and demos but lack durability. Dapr Agents, on the other hand, leverages Dapr workflows to give agents durability. For example, if your cluster shuts down, there’s a network interruption, or a Kubernetes pod goes down, Dapr Agents will spin back up, rehydrate its state, and continue where it left off. You get both AI features and the resiliency you’d expect from non-deterministic workflows.

Swapnil Bhartiya: You mentioned other solutions—what were the pain points or drivers behind the creation of Dapr Agents?

Yaron Schneider: A lot of developers were experimenting with other agentic frameworks. They’d run a POC and then try to take it to pre-production or staging to test how it performed. But as traffic increased or as pods went down, or as they scaled the cluster, they found the agents were disrupted and couldn’t complete execution.

We saw that these frameworks lacked the durability needed to make agentic AI systems reliable. Dapr had those capabilities. A developer from Microsoft saw the same issues and built the first version of Dapr Agents on top of Dapr workflows. He contributed it to the project, and CNCF and the Dapr project announced it about a month ago.

Swapnil Bhartiya: Can you share some use cases you’re seeing with Dapr Agents?

Yaron Schneider: It’s still fairly new, but we already have companies trying it out for various use cases—from DevOps automation (creating an agent that automates Kubernetes clusters), to customer success (an agent that connects to internal data sources and chats with users to solve complex problems), to infrastructure automation (automating CI/CD pipelines). Since it’s a general-purpose framework, you can do pretty much anything with it.

Swapnil Bhartiya: What kind of community is building around Dapr Agents? Is it the same Dapr community?

Yaron Schneider: Yes, it’s the same Dapr community. The developers engaged with Dapr are dealing with the same challenges when it comes to agentic AI systems as developers not yet using Dapr. Use cases span customer success, data processing, observability, and more. Because Dapr Agents is general-purpose, it attracts a broad range of developers.

Swapnil Bhartiya: The beauty of open source is you get instant community feedback. What feedback are you getting, and what improvements are in the pipeline?

Yaron Schneider: Great question. In my opinion, the CNCF ecosystem is still far removed from the AI ecosystem. One goal is to bring those two closer. For example, we’re hearing requests for integration with Anthropic’s MCP Server—a project that brings enterprise data into agentic systems so agents can make better decisions. We think MCP Server integration with Dapr Agents would be an amazing first step toward bringing AI workloads into CNCF projects like Dapr.

Swapnil Bhartiya: What’s next for Dapr Agents?

Yaron Schneider: Beyond MCP integration, we want to extend language support beyond Python to include C# and Java. Python is popular with machine learning experts, but many enterprise application developers come from a C# or Java background. We want to support them as well—that’s definitely coming soon.

Swapnil Bhartiya: Yaron, thank you so much for taking the time to update us on Dapr Agents. As the project progresses, I’d love to chat again. I really appreciate your time.

Yaron Schneider: Thank you so much.