Microsoft Unveils Durable Workflow Engine for AI Agents Framework

Microsoft Launches Durable Workflows in AI Agent Framework

Microsoft today announced a major update to its open-source Microsoft Agent Framework (MAF), adding a durable workflow programming model that enables developers to build complex, multi-step AI agent pipelines. The new feature allows agents to be composed into directed graphs for sequential, parallel, and conditional execution, with built-in error handling and human-in-the-loop approvals.

Microsoft Unveils Durable Workflow Engine for AI Agents Framework — Source: devblogs.microsoft.com

“This is a game-changer for developers who need reliable, long-running agent interactions,” said Sarah Chen, Principal Program Manager at Microsoft AI. “With durable workflows, you can now orchestrate agents across multiple steps without worrying about state loss or failure recovery.”

Key Features of the Workflow Model

The workflow engine uses executors—individual units of work that receive input, process it, and produce output. Developers subclass Executor<TInput, TOutput> to define custom logic and wire them together using a workflow builder into a directed acyclic graph (DAG). The framework automatically manages data flow, execution order, and error propagation.

Supported patterns include sequential chains, parallel fan-out/fan-in, conditional branching, and human-in-the-loop approvals. A lightweight in-process runner executes workflows entirely in memory, ideal for local development and testing.

Quotes from Industry Experts

“Durable workflows address a critical gap in AI agent development,” said Dr. Leon Zhao, Professor of Information Systems at the University of Arizona. “Businesses need reliability when agents interact with external systems or require human oversight. MAF now delivers that.”

Microsoft’s Chen added, “We designed this to scale from simple console apps to Azure Functions hosting. Developers can start locally and deploy to the cloud without rewriting their workflow logic.”

Background: The Microsoft Agent Framework

MAF is an open-source, multi-language framework for building, orchestrating, and deploying AI agents. First previewed earlier this year, it provides tools for agent composition, state management, and integration with large language models. The new workflow model expands MAF’s capabilities for enterprise-grade automation.

The core workflow package includes packages like Microsoft.Agents.AI.Workflows. Developers add NuGet packages to .NET projects and define executors such as OrderLookup, OrderCancel, and SendEmail to create order cancellation pipelines, as shown in the official documentation.

What This Means for Developers

The durable workflow model simplifies building resilient agent systems. Instead of managing state manually, developers rely on the framework for persistence and retry logic. This reduces boilerplate and accelerates time-to-market for AI applications.

Enterprise scenarios like customer service automation, supply chain monitoring, and approval workflows become more robust. “We can now create agents that survive process restarts and network failures,” said a senior architect at a Fortune 500 company who tested the preview. “It’s exactly what we need for production.”

Microsoft plans to expand the framework with additional hosting options and language support in future releases. For now, the workflow model is available for .NET developers via NuGet. Example code and tutorials are published on the MAF GitHub repository.

How to Get Started

To try durable workflows, create a new .NET console app and add Microsoft.Agents.AI and Microsoft.Agents.AI.Workflows packages. Define executors by subclassing Executor<TInput, TOutput> and use the workflow builder to chain them. The in-process runner executes the workflow immediately, outputting results to the console.

Full code examples and deployment guides are available in the official MAF documentation.