Microsoft Foundry for AI-901: Generative AI Apps and Agents Explained
Microsoft Foundry is the heart of the AI-901 exam. Learn the model catalog, grounding/RAG, prompt flow, agents, and the build flow you need to know.
If there's one term you must own for AI-901, it's Microsoft Foundry. The refreshed Azure AI Fundamentals exam puts Foundry at the center of Domain 2, which carries the bulk of the questions. The good news: at the fundamentals level you don't have to build anything — you have to understand what Foundry is for and the steps of shipping a generative AI app or agent. Here's the mental model.
What is Microsoft Foundry?
Microsoft Foundry is Microsoft's unified platform for building, evaluating and deploying generative AI solutions on Azure. Think of it as the workshop where you pick a model, connect your own data, wrap it in an app or agent, test it for quality and safety, then deploy and monitor it — all in one place. For AI-901, that end-to-end story is exactly what's tested.
Generative AI building blocks the exam expects
Models and the model catalog
Foundry's model catalog lets you browse and compare models — large language models like the GPT family, small models like Phi, and open models — then deploy the one that fits your need and budget. Know that bigger isn't always better: small models can be cheaper and faster for narrow tasks.
Prompts, tokens and configuration
You should recognize core knobs from objective 1.2: the prompt (your instruction), tokens (how text is chunked and billed), the system message (sets behavior), and temperature (higher = more creative/variable, lower = more focused).
Grounding and RAG
By default a model only knows its training data. Grounding — commonly via Retrieval-Augmented Generation (RAG) — connects the model to your documents so answers are accurate and current. “Use your own data” on the exam almost always points to RAG/grounding.
Agents
An agent goes beyond chat: it can use tools, call APIs and take multi-step actions to complete a goal. AI-901 expects you to know the difference between a simple chat app and an agent that acts.
The build flow, in plain English
- Pick a model from the catalog.
- Ground it with your data using RAG.
- Orchestrate behavior with prompt flow, tools and (optionally) agents.
- Evaluate quality, groundedness and safety before shipping.
- Deploy and monitor the endpoint — cost, latency and content safety.
Don't forget content safety
Foundry includes safety tooling to filter harmful content and reduce risks like prompt injection. This ties straight back to the responsible AI principles from Domain 1 — the exam loves to connect the two.
Start practicing AI-901 questions for free
Reading about the objectives gets you maybe halfway. The concepts only stick once you answer real questions, get one wrong, read why, and try again. That feedback loop is what actually moves your score on exam day.
Foundry concepts are the highest-yield thing to drill for AI-901. On ExamStudyApp you can run a free AI-901 practice session right now — no account and no credit card required. You'll see Microsoft Azure AI Fundamentals questions one at a time with a full explanation after every answer, so each question doubles as a mini lesson.
When you're ready to go all in, the Microsoft Azure AI Fundamentals (AI-901) practice tests give you unlimited adaptive practice plus timed, full-length mock exams that mirror the real scoring — including a per-objective breakdown so you know exactly what to review before you book the exam.
▶ Try free AI-901 practice questions →
The full AI-901 series
This post is part of our five-part AI-901 study series:
- AI-901 Exam Guide (2026): What's New in Azure AI Fundamentals
- Responsible AI on AI-901: Microsoft's 6 Guiding Principles
- Microsoft Foundry for AI-901: Generative AI Apps & Agents (you are here)
- Azure AI Services on AI-901: Vision, Speech & Document Intelligence
- How to Study for AI-901: A 14-Day Azure AI Fundamentals Plan
