Validates knowledge of building Azure AI solutions, including generative AI, agents, computer vision, text analysis and information extraction.
AI-103 tests whether you can choose, build, secure and run AI solutions in Microsoft Foundry. Questions are scenario-based, so build the pieces yourself in Python, learn when to use each Foundry service, and practise finding details on Microsoft Learn, which is open during the exam.
Example scenario: An app's model calls start failing at peak hours with rate-limit errors. What would you change in the deployment, and what would you monitor?
Example scenario: An agent must look up order status through an internal API and ask a human before issuing refunds over $100. How would you wire the tool and the approval?
Example scenario: A product catalog needs accessible alt text for 10,000 images, with longer descriptions for complex ones. How would you configure the model and output?
Example scenario: A call centre wants each call summarized as JSON with a sentiment score and any account numbers removed. Which features would you combine?
Example scenario: Answers from a RAG app miss facts stored in tables inside scanned PDFs. What would you change in the ingestion pipeline?
AI-103 study guide · Skills measured as of April 16, 2026
Choose the best answer from the options, or the stated number of answers when a question asks for more than one.
An agent must answer questions from a company's policy documents and quote the source. Which pattern should you use?
A. Retrieval brings the relevant passages into the prompt so the answer can cite them. The other options do not supply the source text.
Drag items from a list into the matching places; an item can be used once, more than once or not at all.
Drag each need to the Foundry capability that meets it. The needs are scan invoices for fields, turn speech into text, and block unsafe model output.
Scan invoices for fields → Content Understanding, turn speech into text → Speech to text, block unsafe model output → Content safety filter.
Pick one value from each drop-down list in a block of text, code or a diagram to complete it.
Select the correct value in each list. To avoid storing keys, the app connects to the model by using a [managed identity / API key] and the role [Azure AI User / Owner].
Managed identity and Azure AI User. A managed identity removes stored keys, and the narrow role grants only what the app needs.
Read a description of a fictional company, then answer several questions that share the same background.
Fabrikam wants a support agent that checks order status and hands off to a person for refunds. Which design handles the hand-off?
A. A tool call gets live order data, and an approval step keeps a person in control of refunds.
Complete tasks in a live environment by using the portal or code, and the exam checks the result.
In the project, deploy a chat model and configure the app to call it by using a managed identity.
Create the deployment, assign the app's identity the Azure AI User role on the resource, and call the endpoint without an API key.
Sign in to Microsoft Learn with a personal Microsoft account, complete your certification profile, then schedule AI-103 with Pearson VUE.
Take the free practice assessment to see where you start. Then pair one course with building each skill area in Foundry yourself.
Most candidates spend 50–90 hours. Plan for the low end if you already build AI apps in Python and have used Foundry; plan for the high end if Foundry or agents are new to you.
Use the official learning path for coverage and one video course for structure. Build each area in your own Azure subscription.
Best free official courseDevelop AI apps and agents on Azure (AI-103T00)
Best broad video courseAI-103 - Azure AI Apps and Agents Developer AssociateBest code-along courseAI-103: Azure AI App and Agent Developer - Complete CourseFill the known gaps. Pass reports name topics the learning path covers lightly, such as image inpainting, CLI deployment, endpoint security and prompt shields. Read the documentation for those directly.
Start with Microsoft's free practice assessment on AI Skills Navigator, which needs a sign-in. Then use a larger question bank and rebuild any scenario you got wrong.
Free official checkAI-103 Practice AssessmentLarge question bankAI-103: Azure AI App and Agent Developer - 1000+ Questions
FREE AI-103 Azure AI App and Agent Developer Associate Practice Exams - SamplerChoose the checklist for your test-centre or online appointment.
A detailed June 2026 pass report estimated the official path covers about three quarters of what was tested, naming inpainting, CLI deployment, RBAC and endpoint security, and prompt shields as gaps. A later passer gave similar advice.
Passers describe mostly scenario questions that need a working understanding of how Foundry, agents and SDKs fit together. They recommend hands-on labs and repeating practice questions until the reasoning is clear.
Candidates say Document Intelligence and Content Understanding often appear as competing answers. Learn which fits OCR, layout, charts and complex PDFs. AI-102 experience helped some passers and not others, because AI-103 leans much more on agents.