Microsoft Foundry

AI-103: Developing AI Apps and Agents on Azure

Microsoft Foundry is Microsoft's unified platform for building advanced AI applications and AI agents.

You may have encountered it under its previous name, Azure AI Foundry. Microsoft has consolidated and renamed several of its AI offerings over time, and the current platform is Microsoft Foundry.

The key idea is simple:

While Microsoft Copilot and Copilot Studio provide powerful AI experiences, Foundry is designed for scenarios where developers and AI engineers need significantly more control and customization.

1. What Is Microsoft Foundry?

Microsoft Foundry is designed as a unified AI platform for enterprise applications. It brings together capabilities such as:

• AI models
• AI agents
• Knowledge
• Tools
• Workflows
• Security
• Governance
• Monitoring
• Observability

The goal is to give developers a single platform for building sophisticated AI workloads.

2. Foundry vs Copilot and Copilot Studio

Microsoft provides several ways of working with AI, so it is important to understand the distinction.

Microsoft Copilot

Microsoft Copilot provides ready-made AI assistants for specific scenarios. Examples include:

• Microsoft 365 Copilot
• Security Copilot
• GitHub Copilot
• Azure Copilot

You primarily use these Copilot experiences.

Copilot Studio

Copilot Studio allows you to build your own agents and Copilots using a low-code approach. You can create:

• Agents
• Agent flows
• Workflows
• Connections to data and services

The major advantage is that you can build these experiences without having to write a complete application from scratch.

Microsoft Foundry

Foundry is targeted more toward developers, AI engineers, and organizations that need to build custom AI applications and agents. It provides much greater flexibility for:

• Custom code
• Models
• Agents
• Multi-agent systems
• Workflows
• Tools
• Knowledge
• Enterprise security
• Observability

A useful way to remember the difference is:

3. Why Use Microsoft Foundry?

Consider a simple AI chatbot. You could use an existing Copilot and get a ready-made AI experience.

But suppose your organization wants to build something much more complex. For example, an AI system that:

1. Receives a customer request
2. Understands the request
3. Searches internal knowledge
4. Calls an external API
5. Performs an action
6. Coordinates with another agent
7. Maintains context across interactions
8. Applies security policies
9. Logs and monitors everything

This requires significantly more control. This is where Foundry becomes useful.

4. Custom AI Applications and Agents

Foundry allows developers to create their own AI workloads. Instead of simply configuring an existing assistant, you can build an application around the AI capabilities you need.

This gives developers control over how these components interact.

5. Multi-Agent Orchestration

One of the more advanced scenarios supported by Foundry is multi-agent orchestration. Instead of having one agent perform every task, you can create multiple specialized agents.

Each agent can have its own responsibilities. For example:

• Research Agent - finds relevant information.
• Finance Agent - analyzes financial information.
• Support Agent - handles customer-support tasks.

The orchestrator can coordinate the agents to complete a larger task.

6. Workflows and Orchestration

Foundry can also be used to create complex workflows.

This allows AI applications to combine:

• AI reasoning
• Deterministic workflows
• External APIs
• Enterprise data
• Multiple agents

7. SDKs and Development Frameworks

Foundry isn't limited to a graphical interface. Developers can interact with Foundry programmatically using development kits and frameworks, such as:

• Foundry SDK
• Semantic Kernel
• LangChain

Developers can use these frameworks from their development environment to interact with Foundry capabilities.

This code-first approach is one of the major differences between Foundry and more low-code-oriented AI platforms.

8. Foundry Models

One of the core capabilities of Foundry is the model catalog. Developers can select models that best fit their applications or agents.

The catalog includes models from different providers. Examples include:

• OpenAI
• Meta
• Mistral
• Anthropic

This gives developers the flexibility to choose the model that best fits their requirements.

9. Managed Models, Custom Models and Bring Your Own Models

Foundry can work with different types of models.

Managed Models

These are models made available through Foundry.

The infrastructure for the model is managed as part of the platform.

Custom Models

You can also customize an existing model for your requirements.

For example, you might take an existing foundation model and customize it for a particular business scenario.

Bring Your Own Model

In this scenario, you have a model that you built in your own environment and want to integrate into Foundry.

The important distinction is:

10. Foundry Agent Service

Another major component of Microsoft Foundry is Foundry Agent Service. The purpose of Agent Service is to help developers:

• Build agents
• Deploy agents
• Scale agents
• Orchestrate agents
• Connect agents to tools
• Connect agents to knowledge

A useful analogy is:

You start with an idea or use case and end up with an agent that can be deployed into your organization.

11. Agent Service Architecture

A simplified architecture looks like this:

The agent can then interact with:

• Users
• Knowledge
• APIs
• Enterprise systems
• Other agents

12. Foundry IQ

Foundry IQ acts as a knowledge layer for AI applications and agents. The purpose is to make your applications and agents more knowledgeable by connecting them to organizational information.

This allows an agent to retrieve information from enterprise knowledge sources instead of relying solely on what the underlying model already knows.

Foundry IQ is an intelligent knowledge layer built on top of capabilities such as Azure AI Search.

13. Foundry Tools

Another major component is Foundry Tools. These are Microsoft's pre-built AI APIs for specific use cases. Many of these capabilities existed previously under the name Azure AI Services.

Examples include:

• Vision
• Speech
• Language
• Translator
• Document Intelligence
• Search

Rather than building these capabilities yourself, you can consume them as services.

We'll examine Foundry Tools in more detail in a separate article.

14. Foundry Local

Foundry Local is different from most of the other Foundry capabilities. Instead of running inference entirely in the cloud, Foundry Local provides an option for running supported AI models locally on a device.

There are several ways to interact with Foundry Local:

• CLI
• SDK
• REST API

This can allow a capable device such as a laptop to run supported AI models locally.

15. Knowledge and Tools Make Agents More Powerful

A model by itself can generate responses. But an agent becomes much more useful when it can access knowledge and perform actions.

For example:

• Model - determines what should happen.
• Foundry IQ - provides relevant organizational knowledge.
• Foundry Tools - provide specialized AI capabilities.
• External APIs - allow the agent to interact with business systems.

Together, these components can create a much more capable AI application.

16. External Tools and Services

Foundry agents can also interact with external tools and services. Examples include:

• MCP services
• Azure AI Search
• Azure AI Vision
• SharePoint
• Microsoft Fabric
• Azure Functions

These are examples rather than an exhaustive list.

This is one of the reasons agentic AI can go beyond simple question-and-answer applications.

17. Agent-to-Agent Communication

Modern AI systems can contain multiple specialized agents. Foundry can also integrate with industry-standard agent communication protocols such as A2A and other agent protocols.

This allows agents to communicate and collaborate rather than operating entirely in isolation.

This becomes particularly useful in complex multi-agent architectures.

18. Developer-to-Agent Architecture

Let's put the main pieces together.

This is the central architecture to understand.

19. Security and Governance

Enterprise AI applications need strong security controls. Foundry integrates with the broader Azure security ecosystem, with capabilities such as:

• Role-based access control (RBAC)
• Identity
• Networking policies
• Security
• Monitoring
• Tracing
• Governance

This is important because enterprise AI applications often interact with sensitive organizational data.

20. AI Content Safety

Another important part of an enterprise AI platform is content safety. Foundry provides the ability to incorporate AI safety controls into applications.

This helps organizations build AI applications with appropriate safety controls around user inputs and generated outputs.

21. Observability

AI applications need to be monitored just like other production applications. Foundry provides capabilities around:

• Tracing
• Monitoring
• Performance
• Evaluation
• Observability

For an agent, this can be particularly important because a single user request may involve several operations.

Without observability, it can be difficult to understand why an agent produced a particular result or where a workflow failed.

22. Publishing AI Applications

Another advantage of Foundry is its integration with Microsoft's broader ecosystem. AI applications and agents can be integrated with environments such as:

• Microsoft 365
• Microsoft Teams
• Dynamics
• Other Microsoft business applications
• Custom applications

This makes it possible to take an AI workload developed in Foundry and expose it through the environments where users already work.

23. Why Foundry Is an Enterprise AI Platform

At this point, we can see why Microsoft positions Foundry as more than simply an AI model platform.

A complete enterprise AI solution may require:

Foundry brings these capabilities together.

24. Microsoft Foundry - The Big Picture

The easiest way to visualize Foundry is:

And around the platform you can use developer frameworks such as Python SDKs, Semantic Kernel and LangChain to build custom AI solutions.

25. AI-103 Exam Takeaways

For AI-103, remember these core concepts.

Enterprise Capabilities

Foundry also provides capabilities around:

• Security
• Identity
• RBAC
• Networking
• Governance
• Monitoring
• Observability
• Content safety

Quick Revision

Final Takeaway

Microsoft Foundry is designed to be the unified development platform for advanced AI applications and agents.

The most important architecture to remember is:

Models provide the AI capabilities. Agents use models, instructions, tools, and knowledge to accomplish tasks. Foundry IQ provides the knowledge layer. Foundry Tools provide specialized AI capabilities. Foundry Local provides an option for local inference.

And around all of these, Foundry provides enterprise capabilities such as security, governance, monitoring, and observability.

For AI-103, don't just memorize the Foundry component names. Understand how they fit together:

That architecture is the foundation for understanding the more detailed Deploy Foundry, Pricing, and Foundry Tools lessons that follow.