Training - Orchestrated Workflows

Training content for orchestrated workflows with multiple agents is available in the Microsoft ecosystem, and it aligns closely with what is described as workflows and orchestration workflows in Microsoft Foundry and Azure AI Language.

Key options based on the provided information:

1. Use Microsoft Foundry workflows to orchestrate multiple agents
   • Workflows in Microsoft Foundry are designed to create declarative, predefined sequences of actions that orchestrate agents and business logic in a visual builder.
   • They are ideal when needing to:
     • Orchestrate multiple agents in a repeatable process.
     • Add branching logic (if/else) and variable handling without writing code.
     • Add human-in-the-loop steps such as approvals or clarifying questions.
   • This matches the scenario of building a workflow that coordinates multiple child agents to, for example, gather inputs from a requestor and then generate a competitor list.
   • Documentation includes conceptual guidance and how-tos, including using Visual Studio Code for low-code and pro-code workflows:
     • Work with Declarative (Low-code) Agent workflows in Visual Studio Code.
     • Work with Hosted (Pro-code) Agent workflows in Visual Studio Code.
2. Learn orchestration workflow in Azure AI Language (Foundry tools)
   • Orchestration workflow in Azure Language (accessible via Microsoft Foundry) is specifically intended to route and coordinate different language capabilities (for example, Conversational Language Understanding and Custom Question Answering) within a single project.
   • A custom playground in Microsoft Foundry is provided to iteratively:
     • Tag utterances.
     • Train orchestration models.
     • Evaluate performance before deployment.
   • There is a dedicated quickstart and how-to documentation for creating and training orchestration workflow projects, which can serve as structured self-paced training.
   • Example scenario: an enterprise chatbot that routes requests to different skills (FAQ, calendar operations, feedback processing). This pattern is directly applicable to a multi-agent workflow that, for example, collects user inputs and then routes them to specialized agents (such as data gathering, analysis, and summarization agents) to produce a competitor list.
3. Training the orchestration workflow model (step-by-step)
   • The orchestration workflow documentation includes detailed steps for training models via REST API:
     • Start a training job with a POST request to the orchestration project’s :train endpoint, providing parameters such as modelLabel, trainingMode, and evaluationOptions (training/testing split).
     • Poll the training job status using the operation-location URL until completion.
   • This material functions as hands-on training for building and iterating on an orchestrated language workflow.
4. Concurrent orchestration patterns with the Microsoft Agent Framework
   • The Microsoft Agent Framework documentation for concurrent orchestration shows how to:
     • Define multiple specialized agents (for example, research, marketing, legal).
     • Run them concurrently on the same task using AgentWorkflowBuilder.BuildConcurrent (C#) or ConcurrentBuilder (Python).
     • Collect and aggregate their results.
   • This pattern is well-suited to a scenario where multiple child agents contribute different perspectives to a single request, such as building a competitor list:
     • One agent could research market data.
     • Another could analyze marketing positioning.
     • Another could check legal or regulatory aspects.
   • The documentation walks through setup, execution, and result aggregation, which can be used as a practical guide for building multi-agent orchestrated workflows.
5. How this maps to the requested scenario
   • To build a workflow that creates a competitor list based on several inputs from a requestor:
     • Use Foundry workflows to visually orchestrate the steps: collect user inputs, call multiple agents (for example, data retrieval, analysis, summarization), and combine their outputs.
     • Use the Agent Framework concurrent orchestration pattern if implementing programmatically, so multiple child agents can work in parallel on the same request.
     • Use orchestration workflow in Azure AI Language if the workflow is primarily about routing natural language requests to different language skills (for example, Q&A vs. CLU-based classification).
6. About scheduling or formal training
   • The provided context describes official documentation, quickstarts, concepts, and how-to guides, including a custom playground in Microsoft Foundry for orchestration workflows. These serve as the primary structured learning resources for building orchestrated workflows and multi-agent systems.

References:

• Build a workflow in Microsoft Foundry
• What is orchestration workflow?
• Quickstart: Orchestration workflow (rest-api)
• Train your orchestration workflow model
• Microsoft Agent Framework Workflows Orchestrations - Concurrent (programming-language-csharp)
• Microsoft Agent Framework Workflows Orchestrations - Concurrent (programming-language-python)