Update (September 2026): AI-900 and AI-102 retired on June 30, 2026. Their replacements, AI-901 (Azure AI Fundamentals) and AI-103 (Azure AI App and Agent Developer Associate), are the current exams. For the current path, see the AI-103 Study Guide.
By GetHandsOn.ai Team · April 2026 · 10 min read
If you've been researching the AI-103 exam, you've seen the same word appear in every sentence of Microsoft's official documentation: Foundry.
Not Azure. Not OpenAI. Foundry.
Microsoft has made it unmistakably clear that Azure AI Foundry is not just one feature on the AI-103 exam: it is the platform the entire exam is built around. Every domain, every skill, every scenario-based question you will face assumes you know how Foundry works.
And yet most candidates start their AI-103 prep without ever properly understanding what Foundry actually is.
This post fixes that. By the end you will know exactly what Azure AI Foundry is, how it's structured, what it does, and why Microsoft bet the entire AI-103 certification on it.
What Is Microsoft Azure AI Foundry?
Azure AI Foundry is Microsoft's unified platform for building, managing, deploying, and evaluating AI applications and agents in the cloud.
Think of it as the single workspace where an AI developer does everything: selects a model, configures an agent, connects data sources, sets up safety guardrails, evaluates outputs, and deploys to production, without jumping between five different Azure services.
Before Foundry existed, building an AI application on Azure meant stitching together Azure OpenAI Service, Azure AI Search, Azure Machine Learning, Azure AI Content Safety, and Azure Cognitive Services, all separately, all with different portals, SDKs, and authentication models. It worked, but it was fragmented.
Foundry changes that. It is the unified layer that brings all of those services together under one roof, with a shared project structure, shared security model, and a single SDK (azure-ai-projects) that connects to everything.
Microsoft officially describes Foundry as the platform for model management, agent configuration, and evaluation pipelines. That three-part description maps almost perfectly to the three things you will be tested on in AI-103.
The Foundry Architecture: Hubs, Projects, and Resources
Understanding Foundry's structure is not optional for AI-103. The exam tests it directly.
Foundry organises everything into two levels:
1. The Hub
The Hub is the top-level container in Azure AI Foundry. It holds the shared infrastructure that multiple AI projects can use:
- Azure Storage Account (for data and artefacts)
- Azure Key Vault (for secrets and credentials)
- Azure Container Registry (for model images)
- Azure AI Services connections
- Network and security configuration
You create one Hub per environment (for example, one for your organisation's production AI work). Think of it as the building: everything inside shares the plumbing and utilities.
2. The Project
The Project lives inside a Hub. Each Project represents a specific AI application or workload you're building. Projects share the Hub's infrastructure but have their own:
- Model deployments
- Agent configurations
- Evaluation runs
- Data connections
- Access control
This is where developers actually work. You open a Project, select your model, build your agent, run your evaluations.
Why this matters for the exam: AI-103 scenario questions frequently ask you to identify where something should be configured, at the Hub level or the Project level. Getting this wrong in a real-world deployment causes security and cost problems. Microsoft tests it because it's genuinely important.
Try a real Azure AI lab, free
A pre-provisioned Azure environment in your browser. No Azure account, no credit card, no setup.
Most large enterprises already run on Microsoft infrastructure. Azure AI Foundry is the platform those teams use to build and deploy AI apps, so the skills you practice here map directly to real job demand.
What Can You Do Inside Azure AI Foundry?
Foundry is not just a management layer. It is a fully functional development platform. Here is what it covers:
Model Catalogue and Deployment
Foundry gives you access to a catalogue of models from Microsoft, OpenAI, Meta, Mistral, and others. You browse the catalogue, select a model (GPT-4.1, Llama 3, Phi-4, and more), and deploy it directly to your Project as a managed endpoint.
You can deploy:
- Provisioned throughput deployments for consistent performance at scale
- Pay-as-you-go deployments for development and lighter workloads
- Serverless API deployments for third-party models
For the AI-103 exam, you need to know how to choose the right model for a task: LLMs for language, multimodal models for vision and audio, small models for lower-latency scenarios, and how to deploy them correctly inside Foundry.
The Foundry Agent Service
This is where AI-103 gets serious.
The Foundry Agent Service is Foundry's built-in runtime for AI agents. An agent is an AI system that doesn't just answer a single question. It reasons over multiple steps, uses tools, accesses external systems, and takes actions autonomously.
Inside the Foundry Agent Service, you define:
- Agents: the AI entity with a role, goal, and set of instructions
- Threads: the conversation history for a session
- Runs: the execution of the agent against a thread
- Tools: the capabilities the agent can use, including Azure AI Search, file search, code interpreter, web grounding, function calling, and custom APIs
The AI-103 exam expects you to understand not just what agents are conceptually, but how to configure them correctly in Foundry, including tool schemas, memory, multi-agent orchestration, and when to use the Microsoft Agent Framework (built on Semantic Kernel) for complex multi-agent workflows.
Evaluation Pipelines
One of the most neglected topics in AI-103 prep, and one of the most heavily tested, is evaluation.
Foundry has a built-in evaluation system that lets you measure your AI application's outputs against defined criteria before and after deployment. You can evaluate for:
- Quality (coherence, relevance, fluency, groundedness)
- Safety (content policy violations, harmful outputs)
- Performance (latency, token usage)
The exam tests your ability to design evaluation pipelines, interpret evaluation results, and use those results to make decisions, for example, when to retrain, when to adjust safety filters, when to switch models.
Safety and Responsible AI
Foundry centralises safety configuration. You set up:
- Content filters to block harmful input/output categories
- Prompt shields to detect and mitigate prompt injection attacks
- Groundedness detection to identify hallucinated responses
- Risk and safety evaluations for automated testing of harmful outputs at scale
AI-103 scenarios frequently present you with a situation where something has gone wrong with a deployed AI app and ask you to identify which Foundry safety feature would have prevented it. This is not theoretical. Foundry's safety layer is what separates enterprise AI from experimental AI.
The Foundry SDK
Everything you do in the Foundry portal, you can also do in code. The primary SDK for AI-103 is azure-ai-projects (Python), which gives programmatic access to:
- Creating and managing Foundry projects and deployments
- Building and running agents
- Connecting to AI Search, Document Intelligence, and other services
- Running evaluations
You will also work with azure-ai-inference for direct model inference calls, and the Azure AI Search SDK for RAG pipeline construction.
Azure AI Foundry vs Azure OpenAI Service: What's the Difference?
This is the question that confuses almost every AI-103 candidate.
Azure OpenAI Service is the underlying service that hosts and serves OpenAI models (GPT-4, DALL-E, Whisper, etc.) on Azure infrastructure. It gives you API access to those models with enterprise security, compliance, and network controls.
Azure AI Foundry is the platform layer built on top of Azure OpenAI Service (and many other services). When you use Foundry, you are using Azure OpenAI Service under the hood. You're just doing it through a unified workspace with agents, evaluations, safety, and multi-service orchestration built in.
The simple way to think about it:
| Azure OpenAI Service | Azure AI Foundry | |
|---|---|---|
| What it is | Model hosting service | Unified AI development platform |
| Access | API calls | Portal + SDK + API |
| Models | OpenAI models only | OpenAI + Meta + Mistral + Phi + more |
| Agents | Not built-in | Native Agent Service |
| Evaluations | Manual | Built-in pipeline |
| When to use | Direct API integration | Full AI app / agent development |
For the AI-103 exam: you will use Azure OpenAI Service through Foundry. The exam does not test them as competing alternatives. It tests how you use them together inside a Foundry Project.
Why the AI-103 Exam Is Built Around Foundry
Microsoft did not make Foundry central to AI-103 as a product marketing move. They did it because Foundry represents how enterprise AI development actually works in 2026.
Before Foundry, AI engineers at enterprises were spending as much time managing infrastructure, permissions, and service integrations as they were building actual AI functionality. Foundry exists to eliminate that overhead.
The Microsoft Certified Azure AI Apps and Agents Developer Associate certification validates the skills needed to build and deploy AI apps and agents using production-oriented development practices on Azure by using Foundry, described officially as the unified platform for model management, agent configuration, and evaluation pipelines.
Every domain of the AI-103 exam assumes Foundry is your working environment:
- Planning and managing Azure AI solutions → Hub and Project architecture, security, governance
- Implementing generative AI and agentic solutions → Foundry Agent Service, Microsoft Agent Framework
- Implementing computer vision solutions → Azure Content Understanding in Foundry Tools
- Implementing text analysis solutions → AI Search and Document Intelligence connected through Foundry
- Implementing information extraction solutions → RAG pipelines configured in Foundry
There is no way to pass AI-103 without understanding Foundry. The two are inseparable.
Still reading? Grab the AI-103 exam study guide
One-page reference of every Foundry topic the AI-103 exam tests.
What You Need to Know About Foundry for AI-103 Exam Day
Here is a focused summary of the Foundry knowledge the exam expects:
Architecture
- The difference between a Hub and a Project, and what lives in each
- How to configure secure access using managed identities and private endpoints
- Role-based access control within Foundry
Models
- How to deploy models from the Foundry model catalogue
- How to choose between LLMs, small models, and multimodal models for a given scenario
- Provisioned vs pay-as-you-go vs serverless deployment types
Agents
- The core primitives: Agents, Threads, Runs, Tools
- How to configure function calling and tool use
- Multi-agent orchestration using the Microsoft Agent Framework
Evaluation
- How to set up quality and safety evaluations in Foundry
- How to interpret evaluation metrics and act on them
Safety
- Content filtering configuration
- Prompt shield setup for injection detection
- Groundedness evaluation for RAG outputs
SDK
- The
azure-ai-projectsPython SDK for programmatic Foundry access - How to configure an application to connect to a Foundry project
How to Get Hands-On With Foundry Before the Exam
Reading about Foundry and being able to answer AI-103 scenario questions about Foundry are two different things.
The exam does not test your ability to recall definitions. It presents you with a real-world scenario, a deployed agent returning grounded responses inconsistently, a multi-agent workflow where tool calls are failing, a safety evaluation flagging unexpected content, and asks you to diagnose and fix it.
You can only answer those questions confidently if you have actually built something in Foundry.
That means:
- Creating a Foundry Hub and Project in a real Azure environment
- Deploying a model and making API calls to it
- Configuring an agent with at least one tool (Azure AI Search or function calling)
- Running a basic evaluation against the agent's outputs
- Setting up a content filter and testing it with edge-case inputs
At GetHandsOn.ai, every lab runs inside a real Azure environment. No sandbox simulation, no screenshots to follow. You deploy, configure, and break things in actual Foundry, and the lab guides you through exactly the scenarios the AI-103 exam tests.
If you are preparing for AI-103, start with the free lab: Deploy Your First Azure AI Model with Microsoft Foundry. It takes 30 minutes and gets you hands-on with the exact environment the exam assumes you know.
Summary
Azure AI Foundry is the unified platform Microsoft built for enterprise AI development. It brings together model deployment, agent configuration, evaluation pipelines, safety tooling, and multi-service orchestration into a single workspace.
AI-103 is built around Foundry because Foundry is how real AI applications get built in 2026. Every domain of the exam, planning, generative AI, computer vision, text analysis, information extraction, assumes Foundry is your environment.
The fastest way to be ready for the exam is to spend time inside it.
Ready to commit to AI-103 prep? See pricing and pre-order the full AI-103 hands-on lab bundle and start building inside real Foundry environments today.
Related Reading
- What Is Azure AI Foundry Used For? 8 Things Teams Actually Build With It
- AI-103 Study Guide & Cheat Sheet: All 4 Learning Paths (2026)
- 10 AI-103 Practice Questions: Scenario-Based Exam Prep for 2026
- AI-102 vs AI-103 vs AI-900: Which Microsoft AI Cert Should You Take First in 2026?
- Microsoft AI Certifications Are Changing: AI-900 & AI-102 Retire June 2026
- Why Hands-On Labs Matter More Than Practice Tests for AI-103
Last reviewed September 2026 by the GetHandsOn.ai team.
Confused by the naming? See Foundry vs Azure AI Studio vs Azure OpenAI: what actually changed.