Where are AI-103 practice questions with explanations, not just answers?
I sat down to prepare for AI-103 (Azure AI Apps and Agents Developer Associate) and did what everyone does: searched for practice questions. I found plenty. Udemy has 650-question banks at $9.99. mscertquiz gives 40 free questions with no signup. open-exam-prep has 200 with an AI tutor. Microsoft itself still has no official practice assessment for AI-103.
Every one of them is a question dump. That is the problem.
What AI-103 tests that AI-102 did not
AI-102 tested services: which API does OCR, which one does sentiment, which SKU has which limit. AI-103 tests a platform. The domains are plan and manage, generative AI and agents, computer vision, natural language, and information extraction, and the weight moved hard toward Foundry, agents, RAG and evaluation.
Those topics are mechanisms, not facts. A retrieval score threshold, a token budget, an agent thread that hands a tool result back to the model, a content-safety filter that blocks one prompt and passes another. You cannot memorize a mechanism. You have to watch it move.
Why a question dump fails on mechanisms
A dump trains recognition: you see the right option and it looks familiar. The exam asks for prediction: given this configuration, what happens? Familiarity does not survive a rephrase. Every candidate who "passed all the practice tests" and failed the exam hit exactly this.
The test is simple. Take any practice question you got right and change one number in the scenario. If you cannot say what the new answer is and why, you recognized the question; you did not understand the mechanism.
What I built instead
I built the deck I wanted: 68 slides, 94 interactive widgets, four widget types, and I do not let you read past a concept without doing something with it.
- Steppers. Values change per click. A token budget fills, a retrieval score drops below the floor, an agent thread grows one message at a time.
- Hotspots. Click a box in an architecture and see what it does and where it fails.
- Sliders. Move temperature, top-p, chunk size or k and watch the output change.
- Drills. Guess before reveal, at the end of every domain.
Then a 40-question simulator with an explanation for every answer, not just the right one, with progress saved in the browser. Every question is original, written from Microsoft's published skills outline. Nothing reproduces live exam content; that is the exam NDA and I take it seriously.
It ships as a zip: open index.html in Chrome or Edge, works offline, plus a one-page infographic of the blueprint and a printable PDF deck.
AI-103 Flight School — $49 on Whop · details on the kits page.
If you only do one thing
Take the official skills outline, and for each line ask "what moves here?" If the answer is a number or a state, find something that lets you move it before you memorize anything about it.
Microsoft, Azure and Microsoft Foundry are trademarks of the Microsoft group of companies. Flight School is an independent study aid, not affiliated with, sponsored by or endorsed by Microsoft.
Related notes
- GenAI interview questions with model answers: what the interviewer is actually probing
46 GenAI interview questions with model answers and 45 troubleshooting cards: what AI-engineer panels ask about RAG, agents, LLMOps and fine-tuning. - The five rungs of agent engineering
Prompt, context, loop, graph, ownership. Most teams stall on rung one and wonder why their AI never leaves the demo. A ladder for what to fix next. - Make your agent fail closed
The worst agent bug isn't a wrong answer — it's a confident answer built on a tool that quietly returned nothing. A design rule, and where to enforce it.