Tips for the Azure AI Engineer Associate exam.

May 20264 min read

Annoyingly, I passed AI-102 just before Microsoft retired it. Its replacement, AI-103, means I have to sit another one. Take these tips from someone who's just finished AI-102 and is starting AI-103.

Sitting AI-102 was interesting, because going through it you could see why Microsoft was about to retire it. The questions were heavy on Document Intelligence, the Language service, and Custom Vision. All useful. But the exam tests how to use those services on their own, not how to wire them into anything bigger. AI-103 rebuilds the cert around Azure AI Foundry, agents, and multi-agent orchestration. That's the work AI engineers actually do now.

Do practice exams, then do them again

Microsoft Learn modules teach you the services. The exam asks you to pick the right one in 90 seconds. Different problem.

Do a full practice set. Mark every wrong answer. Mark every slow answer too. A week later, redo only the marked ones. You're ready when you can clear 85% twice in a row without rushing.

If you're going to pay for one thing, make it a MeasureUp practice set for AI-102 (about $80 USD). Free practice tests get you most of the way. The question style and tone on MeasureUp are far closer to the real exam than anything else I tried.

Run the Python labs on GitHub

There's a Python labs repo for AI-102 that runs against real Azure services. Free.

Open it in GitHub Codespaces (one click from the repo) and you get a full VS Code environment in the browser, no local setup. The labs are Jupyter notebooks: a mix of explanation and runnable Python cells. You read a paragraph, hit Run on the code below it, and watch the service respond. Six notebooks cover Python foundations, data structures, functions and error handling, files and JSON, REST APIs, and an end-to-end AI-102 pattern.

Screenshot of the python-for-ai102 repo open in GitHub Codespaces, showing a Jupyter notebook with Python code for working with Azure AI Services response data. — The repo open in Codespaces, lab 2 (data structures). Read the prose, hit Run on the cell below it.

Far more useful than reading the docs. You stop treating service names as words on a page and start treating them as things with a shape.

If you only read modules, you'll pass nothing.

Build something on the side

While you're studying, build something with the services you're being tested on. I had a side project on the go while studying for AI-102: Azure AI Foundry, Azure AI Search, a small RAG system on top. None of it was production-grade. All of it taught me more than the modules did.

Pick anything that interests you. The exam covers most of the useful Azure AI services, so whatever you build will touch three or four of them. The modules suddenly mean something.

Flag and move

130 minutes, 40 to 60 questions. If a question takes more than 90 seconds, flag it and come back. Case studies will eat your time if you let them.

Good luck. matt@matthitchman.com if you've sat either of them.

Matt