Microsoft Azure AI Apps and Agents Developer Associate (AI-103)

About the AI-103 exam

AI-103, Developing AI Apps and Agents on Azure, is the exam for the Microsoft Certified: Azure AI Apps and Agents Developer Associate certification. It is aimed at developers who build generative AI applications and agents on Microsoft Foundry and write the code in Python. Microsoft expects you to know general and generative AI concepts and the Azure services around them, not just how to call a chat model.

The questions are mostly practical design and build decisions. You are given an app or an agent with a requirement (cheaper inference, grounded answers, safer output, a document that must turn into structured fields) and you pick the Foundry feature, deployment type, tool or setting that meets it. Wrong options are often real features that solve a neighbouring problem: a content filter where a blocklist was needed, function calling where an OpenAPI tool fits better, or a standard deployment where batch would cost far less.

Exam format

• Questions: Microsoft doesn't publish an exact number; it says most of its certification exams have 40 to 60. Expect multiple choice plus other item types such as case studies, drag-and-drop and hot area, and possibly interactive components.
• Time: 120 minutes.
• Passing score: 700 on a scaled score out of 1,000.
• Cost and delivery: the price depends on the country where you take it. It is proctored and scheduled through Pearson VUE.
• Languages: English, Chinese (Simplified and Traditional), French, German, Japanese, Korean, Italian, Portuguese (Brazil) and Spanish.
• Validity: 1 year. You renew for free by passing an online assessment on Microsoft Learn.
• Open book, partly: role-based exams let you browse Microsoft Learn during the exam, but the clock keeps running, so it only helps with the odd lookup.

What the exam covers

The skills outline, current as of April 16, 2026, has five domains:

• Plan and manage an Azure AI solution (25-30%). Choosing models and Foundry services, deployment options, quotas and rate limits, cost, managed identity and keyless access, private networking, and responsible AI controls such as content filters, guardrails and evaluations.
• Implement generative AI and agentic solutions (30-35%). RAG in an application, agents with tools, function calling and memory, multi-agent orchestration, approval flows, prompt and parameter tuning, tracing and token analytics.
• Implement computer vision solutions (10-15%). Image and video generation and editing, including mask-based edits, captions and alt text from multimodal models, Azure Content Understanding for images and video, and blocking unsafe visual content or prompt injection hidden in images.
• Implement text analysis solutions (10-15%). Entity, sentiment and PII extraction, JSON output from prompts, translation with Azure Translator, and speech to text and text to speech as an agent modality.
• Implement information extraction solutions (10-15%). Indexing content in Azure AI Search with semantic, hybrid and vector search, enrichment skills and OCR, and turning documents into structured or Markdown output with Content Understanding.

How the practice questions here are organised

There are 10 practice tests of 10 questions each, and every test covers all five domains, with planning and agent questions taking the larger share as they do on the real exam. Topics range from 429 throttling and batch deployments to agent tracing, PII redaction in conversations and document field extraction. Each question explains why the right answer works and why each wrong option doesn't, with a link to the Microsoft Learn page it is based on.

The full practice exam has 50 questions in 120 minutes, split across the domains by their weights, with answers hidden until you submit. Microsoft doesn't publish how its scaled score maps to correct answers, so the mock passes you at 70%. Case studies and drag-and-drop items are asked here as multiple choice or "choose two" questions.

A study plan that works

• Take Test 1 before studying to see which domains you're weakest in.
• Build something small in Foundry: deploy a model, add a grounding index, then wrap it in an agent with one tool. Many questions assume you've seen these screens and SDK calls.
• Read the explanations of the questions you missed or guessed, then follow the linked docs. Foundry features change names and move often, so trust current documentation over old notes.
• Spend extra time on the two big domains, since together they are well over half the exam.
• When most tests pass comfortably, sit the full practice exam timed and in one go, and try the exam sandbox so the item types aren't a surprise on the day.

These questions are original, written from the public Microsoft Learn documentation and the official study guide. They are not real exam questions, and this site is not affiliated with Microsoft.