Hi, I’m Lena Nadir.
Microsoft released “AI-103: Azure AI Apps and Agents Developer Associate” back in July 2026. It’s a certification that covers building generative AI applications and autonomous AI agents, but being such a new cert comes with a specific problem: it’s genuinely hard to find hands-on material that matches the actual difficulty of the exam.
Microsoft Learn is a solid primary source, but reading through the docs alone doesn’t always make it clear how to actually turn that into working code and wire everything together.
So on this blog, drawing on my own experience passing AI-103 on the first attempt, I’m starting a hands-on series. AI-103 covers five domains in total, but this series focuses specifically on the Generative AI/Agent domain, which carries the highest weighting on the exam. I picked these themes with both exam prep and real-world usefulness in mind, not just “studying for the test.”
This post is the index for four hands-on articles I’ll be publishing. I’ll add links below as each one goes live.
🛠️ Four Hands-On Themes, Focused on Generative AI/Agent
This series zeroes in on the Generative AI/Agent domain, the highest-weighted part of AI-103, starting with the fundamentals and building up to production-level depth. I’ve kept the scope to four themes you can work through one step at a time, without jumping straight into advanced territory.
Part 1: Defining an AI Agent in Microsoft Foundry Portal and the Basics of the Python SDK
- What you’ll do: Start by defining your first AI agent no-code in the Foundry portal and verifying it works. From there, use Python to call that agent through the SDK and build a basic agent flow that runs a simple task.
Part 2: Implementing Basic RAG with File Search
- What you’ll do: This is the first step toward giving your agent its own knowledge base. Using the File Search tool in Microsoft Foundry, you’ll upload documents and implement a RAG setup where the agent references those files to answer accurately, via the Python SDK.
Part 3: Getting JSON Structured Output
- What you’ll do: Controlling an LLM’s response into a format your program can actually work with is a must-have skill for app development. You’ll use Structured Outputs, an important item on the AI-103 syllabus, defining a schema to reliably get output in JSON.
Part 4: Implementing Responsible AI on Azure
- What you’ll do: To build a safe and trustworthy AI system, you’ll learn how to configure agent guardrails. This covers blocking harmful content, defending against prompt injection, and setting up intervention points — the Responsible AI principles that show up often on the exam.
AI-103 also includes domains like Computer Vision, Text Analysis, and Information Extraction, but those are out of scope for this series.
🚀 How This Series Works, and What Comes Next
Links to each hands-on article will be added below as they’re published.
- Part 1: Defining an AI Agent in Microsoft Foundry Portal and the Basics of the Python SDK
- Part 2: Implementing Basic RAG with File Search
- Part 3: Getting JSON Structured Output
- Part 4: Implementing Responsible AI on Azure (link coming soon)
Work through each article in your own Azure environment as you read. Actually getting hands-on will get the concepts to stick in a way that just reading documentation won’t.
💡 Want to Check Your Readiness Before Exam Day?
Once you’ve worked through the hands-on material, if you want to check whether your understanding is actually at exam level, I’ve published an AI-103 practice test collection on Udemy.
It doesn’t just mark answers right or wrong — it explains why the other options are incorrect, grounded in the official specs. Combined with the hands-on series, it should make you more comfortable with the actual question formats on exam day.
🎟 Launch coupon: RELEASE-CAMPAIGN (33% off, through 2026/09/24) → Apply the coupon here
See you in Part 1.