AI-103 exam prep: practice tests, labs and flashcards.

The Azure AI Apps and Agents Developer exam is Python-first, Foundry-first and heavy on agents. The practice tests match the real question shapes, including code completion. The hands-on labs run in your own Azure subscription, and the labs and flashcards are free.

Take the free AI-103 sample Start the labs (free)

$9.99 a month unlocks every AI-103 and AZ-104 practice test. Cancel anytime.

Code completion, from the free sample

The app must sign in as the user-assigned identity id-foundry, with no secrets stored.

credential = DefaultAzureCredential(
    managed_identity_client_id=
        ID_FOUNDRY_CLIENT_ID
)
project = AIProjectClient(
    endpoint=PROJECT_ENDPOINT,
    credential=credential,
)
Why: managed_identity_client_id picks the user-assigned identity, and it takes the client ID, not the principal ID. Without it, the credential uses the system-assigned identity, which has no role.

What the exam is

AI-103: Microsoft Certified Azure AI Apps and Agents Developer Associate went live in June 2026 and retired AI-102 the same month, with no bridge exam. It measures whether you can plan, build, secure, and operate AI solutions on Microsoft Foundry: generative AI apps with the Responses API, agents with tools, knowledge and MCP servers, multi-agent orchestration with the Agent Framework, information extraction with Azure AI Search, Document Intelligence and Content Understanding, computer vision including image and video generation, and language, speech, and translation services. It is the prerequisite for the expert-level AI-500.

Two things set it apart from most Azure exams. It is Python-first: roughly a quarter to a third of the questions are SDK code completion, drag-and-drop, or JSON schema items, and passers say the code questions are Python only. And the Foundry surface moves quickly, so the study guide (last revised April 2026) and the official exercises change more often than AZ-104's ever did. Everything on this page is checked weekly against the study guide, the MicrosoftLearning exercise repos, and community exam reports.

60
questions: 51-53 standalone, a 7-question case study, 3 yes/no
120
minutes, with Microsoft Learn open during the exam
700
out of 1000 to pass
~25%
Python code completion, drag-and-drop and JSON items

AI-103 exam domains

Plan and manage an Azure AI solution 25–30%
Choosing the right Foundry resource, deployment types and quota, RBAC and keyless auth with managed identity, private networking, guardrails, monitoring and cost, CI/CD and containers.
Generative AI and agentic solutions 30–35%
Responses API and the Foundry and OpenAI SDKs, tools and function calling, prompt engineering, RAG and fine-tuning, Foundry Agent Service, custom and MCP tools, Foundry IQ, workflows, Agent Framework, multi-agent patterns, A2A, and agent observability.
Information extraction 10–15%
Azure AI Search indexes, indexers, skillsets and vector search; Document Intelligence prebuilt and custom models; Content Understanding analyzers; RAG ingestion pipelines.
Computer vision 10–15%
Vision-enabled chat, image analysis with Content Understanding, video analysis, image generation and editing (inpainting), and video generation with Sora.
Text analysis, speech, and translation 10–15%
Azure Language (entities, sentiment, PII, summarization), the Language and Speech MCP servers, speech recognition and synthesis, Voice Live agents, and Translator.

AI-103 practice tests and Domain Focus drills

Five full-length tests and five domain drills, written to the April 2026 study guide. Start with the free sample; the All-Access subscription unlocks every test and drill.

AI-103

Azure AI Apps and Agents Developer Associate

Five full-length tests (60 questions with a case study, timed at 120 minutes) and five 30-question Domain Focus drills, all with code-completion items and full explanations.

All AI-103 practice tests
  1. Loading tests...

Free labs and flashcards

Both are live and free, with no subscription needed.

37 hands-on labs

One lab per course day plus six scenario labs with planted faults. The extension deploys a Foundry resource into your subscription, walks you through the portal and the Python side by side, and verifies each outcome against live Azure: keyless auth, content filters, deployment types, search connections, tracing, and more.

See the AI-103 lab catalog

100 flashcards

Twenty cards per domain covering the vocabulary and the classic exam pairs: Standard vs. Global Standard, function tool vs. MCP vs. OpenAPI, RAG vs. fine-tuning, Document Intelligence vs. Content Understanding, keys vs. managed identity. Spaced repetition, no account needed.

Start reviewing

The free course: 32 days, foundations first, then by exam weight

Each day pairs a teaching video with a lab video. Days marked as gap days cover topics the official learning path underserves, according to people who passed.

  1. Days 1-7: Foundations (25-30%)

    What Foundry is, tools and SDKs, deploying and securing AI resources (gap), model catalog and deployment types (gap), guardrails and Prompt Shields (gap), evaluation and monitoring, CI/CD and containers (gap).

  2. Days 8-20: Generative AI & Agents (30-35%)

    Chat apps and the Responses API, tools, optimization and fine-tuning, Foundry Agent Service, custom and MCP tools, Foundry IQ, Teams publishing, workflows, Agent Framework, multi-agent patterns, A2A, and agent operations (gap).

  3. Days 21-26: Extraction & Vision (20-30%)

    Azure AI Search and knowledge mining, Document Intelligence, Content Understanding, DI vs. CU and RAG ingestion (gap), vision-enabled chat, image and video generation with inpainting (gap).

  4. Days 27-32: Language, Speech & Exam Prep

    Azure Language and the Language MCP server, Speech, Voice Live agents, Translator, then question-type walkthroughs, the case study, and a timed rebuild from an empty subscription to a published agent.

A few of the labs

Every lab ends on code. The full catalog, install instructions and progress sync are on the Labs page.

Day 3Lock Down the Foundry ResourceRetire the keys, give a managed identity the Azure AI User role, restrict the endpoint to your IP, and share one connection across two projects.
Day 5Raise the GuardrailsBuild a custom content filter with Prompt Shields, attach it to a deployment, and call Content Safety from Python.
Day 14Ground the Agent in Your DocsCreate an AI Search service, switch it to Entra auth, connect it to the project, and build a Foundry IQ knowledge base an agent cites.
Day 20Trace, Evaluate, and Identify an AgentWire Application Insights tracing, diagnostic logs, and a user-assigned identity so an agent reaches its tools without a key.
ScenarioRate Limit TriageA chatbot returns 429s at lunch. Read the deployment, raise capacity, switch to Global Standard, and add a budget and alert.
Day 32The Sixty-Minute RebuildFrom an empty subscription to a keyless, grounded, traced agent, against the clock.

AI-103 questions

How is the AI-103 exam structured?

About 60 questions in 120 minutes: 51 to 53 standalone questions, one seven-question case study, and three yes/no questions at the end that cannot be reviewed. Roughly a quarter of the questions are Python SDK code completion, drag-and-drop, or JSON schema items. The passing score is 700 out of 1000, and Microsoft Learn is available during the exam.

Is Microsoft Learn enough to pass?

Community reviews estimate the official learning path covers about 75% of what is asked. Topics that people who passed say were underserved include deployment types and quota, RBAC and keyless authentication, sharing resources between Foundry projects, Prompt Shields, CI/CD pipelines and containers, Document Intelligence versus Content Understanding, and AI Search index schemas. The course plan gives those topics their own days and labs.

Do I need to know Python?

Yes. Microsoft lists Python as the prerequisite language and passers confirm the SDK questions are Python only. You do not need to be an expert: the questions test whether you recognize the right class, parameter, or call order. Every lab's code card and the flashcard deck drill exactly those shapes.

Do the labs cost money to run?

The labs run in your own Azure subscription. Foundry resources are free to create; model deployments bill per token, usually cents per lab, and need model quota in your region. A few labs use resources that bill hourly (Azure AI Search on the Basic tier, for example) and are labeled so you can clean up the same day. The extension deletes everything with one button.

I hold AI-102. Do I need AI-103?

AI-102 retired on June 30, 2026 with no bridge exam, so renewal means sitting AI-103. Passers who came from AI-102 disagree on the overlap: the classic services (Language, Speech, Document Intelligence, Search) carry over, but the agents block (Foundry Agent Service, MCP, Foundry IQ, Agent Framework, workflows) is new and is the largest domain.