Microsoft Foundry Full Tutorial: From Basics to Agentic AI (2026)

Summary

Microsoft Foundry is presented as Microsoft’s unified, model-agnostic platform for building, grounding, deploying, monitoring, and governing AI agents. The tutorial explains that Microsoft consolidated earlier Azure AI tools into one platform to simplify development, support the transition from prototypes to production, and provide access to a broad range of models.

Key concepts and platform comparison

• Agent lifecycle: Foundry supports three stages:
  1. Build — Select a model, then add instructions, tools, knowledge, memory, and guardrails.
  2. Deploy — Make agents available to users and connect them to channels and identity and authentication systems.
  3. Operate — Trace agent runs, evaluate quality, monitor performance, and govern the system. The presenter emphasizes that operating and evaluating agents is especially important because their outputs can be nondeterministic.
• Model choice: The interface offers models from providers including OpenAI, Anthropic, and Grok. The presenter describes the catalog as containing more than 1,900 models. Model parameters include output format, tool-use behavior, and reasoning effort, which can affect cost.
• Microsoft AI tools:
  • Microsoft 365 Copilot is framed primarily as a user-facing productivity and consumption experience.
  • Copilot Studio is the low-code option, with a simpler visual interface aimed at makers.
  • Foundry gives developers more granular control over models, configuration, deployment, and observability.

The presenter generally recommends starting with Copilot Studio when simplicity is the priority and moving to Foundry when deeper control is needed.

• Why use Foundry: The presenter highlights its Azure integration, centralized governance and observability, Microsoft ecosystem connections, and model flexibility, compared with assembling and maintaining integrations across separate tools.

Interface tour and tutorial

The walkthrough starts at ai.azure.com, switches to the newer Foundry experience, and creates a project. Project setup includes an Azure resource group, subscription, and region. The presenter cautions that regional availability can limit which capabilities are usable.

The agent playground includes:

• A model selector and configuration parameters.
• Optional voice mode, with speech model, language, and voice settings.
• Instructions, tools, knowledge, memory, and guardrails.
• Chat for testing, YAML for viewing the agent configuration, and Core agent options for calling it from custom interfaces.
• Details, traces, monitoring, and evaluations for version information, debugging, performance tracking, and test-script-based assessment.
• A broader navigation panel for deployments, fine-tuning, tools, knowledge, memory, evaluations, and guardrails.

The demo builds a Microsoft documentation assistant:

1. The agent is instructed to answer as a Power Platform expert, use only the supplied documents, and say when those documents do not contain an answer.
2. Several documents are uploaded and indexed. The presenter explains that a vector index lets AI search by meaning rather than only by exact words.
3. The agent answers a question using those documents and provides citations.
4. A Microsoft Learn MCP server is added as a tool so the agent can retrieve current information and links. The presenter demonstrates configuring tool approvals and advises matching approval settings to the project’s needs.

Workflows and lifecycle note

The tutorial also demonstrates a visual workflow: one agent answers a question using the documentation, and a second agent reformats the answer as bullet points. The workflow canvas is compared to Power Automate, with nodes for questions, messages, loops, values, and invoking agents. Runs can also be traced, monitored, and evaluated.

In a later recorded note, the presenter says Foundry’s visual workflow designer is still available at the time of recording but is scheduled for retirement on December 1, 2026, with Microsoft moving toward a code-first approach. He advises viewers to consider that transition when designing solutions.

The video does not provide a detailed treatment of pricing, SDK usage, or deployment implementation; these are identified as topics for a separate, deeper guide.

Main speaker and sources

• Halden, identified in the introduction as a Microsoft MVP, Microsoft Certified Trainer, and founder of Power Academy. A brief later insert refers to him as “Holden,” likely an auto-caption inconsistency.
• Microsoft documentation, referenced by the presenter for information about the workflow-designer retirement and migration.