AI-103 Azure AI Apps and Agents Developer Associate Study Guide
AI-103 validates your ability to build, manage and deploy AI apps and agents on Azure using Microsoft Foundry. This AI-103 study guide sets out the exam domains, the format and a preparation plan, so your study time goes on what the exam measures.
What the AI-103 Exam Tests
AI-103 targets an Azure AI engineer who builds, manages and deploys agents and AI solutions using Microsoft Foundry. You should have experience developing apps in Python, and be familiar with general AI, generative AI and Azure services.
The role covers five areas of work:
- Planning and managing Azure AI solutions
- Implementing generative AI and agentic solutions
- Implementing computer vision solutions
- Implementing text analysis solutions
- Implementing information extraction solutions
You work with business stakeholders, solution architects, data scientists, DevOps engineers and cloud security engineers to design, build and maintain AI solutions. AI-102 split generative AI (15-20%) and agentic solutions (5-10%) into two domains. AI-103 combines them into one domain at 30-35%, the largest weighting on the exam.
Target Audience and Prerequisites
The certification is pitched at an intermediate level for two overlapping roles: AI Engineer and Developer. You should have:
- Experience developing applications in Python
- Familiarity with the capabilities of general AI and generative AI
- Familiarity with Azure services
- A working knowledge of Microsoft Foundry, because the exam assesses deployments, agent workflows and model configuration through it
If Azure AI is new to you, work through the Microsoft Foundry documentation and training modules before you start timed practice. Several exam objectives assume you already know your way around a Foundry project.
Exam Format and Scoring
AI-103 follows Microsoft's standard proctored certification format:
- Duration: 120 minutes
- Passing score: 700 or greater, on a scale of 1 to 1,000
- Delivery: Proctored, scheduled through Pearson VUE
- Languages: Offered in English and 9 other languages, including Chinese, French, German and Japanese
- Retakes: You can retake the exam 24 hours after a first failed attempt; the wait for subsequent retakes varies
Most questions cover generally available features. The exam may also include questions on preview features that are already in common use. Many of the skills measured start with "choose", so practise picking the Foundry service, model or configuration that fits a stated business requirement.
Domain Weightings and Blueprint
The current AI-103 blueprint (the "skills measured as of April 16, 2026" version) has five domains:
| Domain | Weight |
|---|---|
| Plan and manage an Azure AI solution | 25-30% |
| Implement generative AI and agentic solutions | 30-35% |
| Implement computer vision solutions | 10-15% |
| Implement text analysis solutions | 10-15% |
| Implement information extraction solutions | 10-15% |
Plan and Manage an Azure AI Solution
This domain is about choosing and running Foundry services. It involves little application code.
Choosing Foundry services: know how to pick a model for a task, including large language models, small language models, multimodal models and Foundry Tools. You also need to match a service to generative tasks, grounding, vector search, agent workflows or multimodal processing, and pick a retrieval and indexing method.
Setting up AI solutions: design Azure infrastructure for AI apps and agent-based solutions, choose deployment options, configure model and agent deployments, and integrate Foundry projects with CI/CD pipelines.
Managing, monitoring and securing: manage quotas, scaling, rate limits and cost footprints for model and agent workloads. Monitor model performance, drift, safety events and grounding quality, along with data ingestion quality, search index health and relevance. Configure security using managed identity, private networking, keyless credentials and role policies.
Responsible AI: configure safety filters, guardrails, risk detection and content moderation. Apply responsible AI instrumentation such as evaluators and safety evaluations, implement auditing through trace logging and provenance metadata, and govern agent behaviour with oversight modes, constraints and tool-access controls.
Implement Generative AI and Agentic Solutions
This is the largest domain on the exam, and the one that changed most from AI-102.
Building generative applications: deploy and consume large language models, small models, code models and multimodal models. Implement retrieval-augmented generation in an application, and design workflows, tool-augmented flows and multistep reasoning pipelines. Evaluate models and apps for fabrications, relevance, quality and safety, and integrate generative workflows using Foundry SDKs and connectors.
Building agents: define agent roles, goals, conversation-tracking approach and tool schemas. Build agents that integrate retrieval, function-calling and conversation memory, and integrate agent tools such as APIs, knowledge stores, search, content understanding and custom functions. Implement orchestrated multi-agent solutions, build autonomous or semiautonomous workflows with safeguards and approval controls, and integrate monitoring so you can evaluate agent behaviour and perform error analysis.
Optimising and operationalising: tune generation behaviour through prompt engineering and model parameters, implement model reflection and self-critique loops, set up observability through tracing, token analytics, safety signals and latency breakdowns, and orchestrate multiple models or hybrid LLM and rules-based flows.
Example AI-103 Scenario
The scenario below tests a skill from the generative AI and agentic solutions domain, the heaviest on the exam.
A retail company is building a customer support agent in Foundry. The agent needs to look up order status from an internal API, answer product questions grounded in a document repository, and remember earlier turns in the same conversation. Which combination of capabilities should the agent use?
- A. A single large language model call with the entire document repository pasted into the prompt each turn
- B. Function-calling to the order API, retrieval-augmented generation against the document repository, and conversation memory
- C. A small language model fine-tuned once on the document repository, with no live API access
- D. A separate chatbot for each task, with no shared context between them
B is correct. The scenario needs three capabilities AI-103 tests separately: function-calling for the live order lookup, retrieval-augmented generation for grounded product answers, and conversation memory to track earlier turns. Option A does not scale past a small repository, since it resends the same content every turn. Option C cannot answer questions about documents added or changed after the fine-tuning run. Option D drops context between tasks, so it fails the conversation-tracking requirement that Foundry agents are built around.
Implement Computer Vision Solutions
The vision domain covers generating images and video as well as understanding them.
Image and video generation: implement solutions that generate images and videos from text prompts and reference media, configure image-editing workflows including inpainting, mask-based edits and prompt-driven modifications, and edit generated videos using the platform's generation and editing controls.
Multimodal understanding: build solutions that analyse visual context using multimodal models, configure concise or detailed captions for single or multiple images, and implement question-answering grounded in visual evidence. Configure alt-text and extended image descriptions aligned to accessibility guidelines, implement visual understanding through Azure Content Understanding in Foundry Tools, and process video segments using single-task and pro-mode Content Understanding pipelines.
Responsible AI for multimodal content: implement filters to classify unsafe or disallowed visual content, detect and mitigate indirect prompt injection embedded as text in images, and enforce visual policy rules such as watermarks and brand usage requirements.
Implement Text Analysis Solutions
Language model text analysis: implement solutions to extract entities, topics, summaries and structured JSON output using generative prompting and Foundry Tools, configure detection of sentiment, tone, safety issues and sensitive content, and build translation solutions using Azure Translator in Foundry Tools or LLM-powered translation flows.
Speech solutions: implement speech-to-text and text-to-speech workflows for agentic interactions, integrate speech as an agent modality including custom speech models, enable multimodal reasoning from audio inputs, and translate speech into other languages using language models and Foundry Tools.
Implement Information Extraction Solutions
Retrieval and grounding pipelines: ingest and index documents, images, audio and video. Configure semantic search, hybrid search and vector search for grounding, implement enrichment using custom or built-in skills for text, images and layout, configure RAG ingestion flow including OCR, and connect retrieval pipelines directly to workflows and agent tools.
Extracting content from documents: extract information using multimodal pipelines that combine OCR, layout analysis and field extraction, produce clean, grounded representations for agents and RAG using Content Understanding, and implement analyzers that generate structured or markdown output for downstream reasoning.
Study Timeline and Preparation Strategy
Plan on 8 to 10 weeks of steady study for AI-103, and longer if Microsoft Foundry is new to you.
Weeks 1-2: Foundation. Work through the Foundry documentation and training modules for each domain. Provision a Foundry project and deploy a model so you have a working environment to practise in.
Weeks 3-5: Generative AI and agents. The exam weighting is heaviest here, so these weeks get proportionally more time. Build a retrieval-augmented generation application end to end, then build an agent with function-calling, tool integration and conversation memory. Practise orchestrating more than one agent in a single workflow.
Weeks 6-7: Vision, text and extraction. Implement an image or video generation workflow, a document extraction pipeline using Content Understanding, and a text analysis solution that produces structured output. Configure responsible AI controls, such as content filters, on at least one of these.
Weeks 8-10: Practice tests and review. Take full-length practice tests to find weak domains, then revisit the Foundry documentation for anything you scored poorly on. Review scenario questions that ask you to choose between similar services, because that pattern turns up in every domain.
Common Exam Traps and How to Avoid Them
Service selection under Foundry. Several questions describe a scenario and ask which Foundry service or model fits. The trap is usually two services with overlapping capability, such as a general multimodal model versus a purpose-built Content Understanding pipeline. Look for clues about whether the scenario needs structured, grounded output or open-ended generation.
Treating agents as a single-step feature. The exam separates agent definition, tool integration, orchestration and monitoring into distinct skills. Calling one function once and running an orchestrated, monitored multi-agent workflow are different skills, and the skills list names each one.
Skipping responsible AI until the end. Responsible AI controls appear inside the planning domain and again inside the computer vision domain. Build guardrails, content filters and prompt-injection mitigation into your practice projects from the start. Treating them as an afterthought is the mistake that costs the most marks here.
Under-preparing for extraction pipelines. Its weighting makes document and multimodal extraction look like a smaller, easier domain. Scenario questions here combine OCR, layout analysis and Content Understanding, and reading alone will not prepare you for them. Build at least one pipeline yourself.
Essential Study Resources
The AI-103 study guide on Microsoft Learn is the primary reference for the domains, weightings and update history above.
Training paths and modules cover Foundry, generative AI, agents, computer vision, text analysis and information extraction, each with interactive exercises and knowledge checks.
The exam sandbox lets you try the question types and interface before exam day.
AI Skills Navigator hosts the official practice assessment for this certification; you need to be signed in to launch it.
Microsoft Foundry documentation is where you will spend most of your hands-on time, since almost every domain routes through a Foundry project, deployment or tool.
azureprep.com has practice questions across current Azure certifications, for working through exam-style questions alongside the official resources.
Taking the Exam
Schedule AI-103 through Pearson VUE from the certification page. Microsoft recommends registering with a personal MSA account. A work or school account puts your exam records at risk: they are lost if you leave that organisation.
The exam is proctored and may include interactive components, so review Microsoft's exam duration and experience guidance before your session. Read each scenario in full before you look at the options, work out which requirement the question is testing, and pace yourself across the 120 minutes so there is time left to return to flagged questions.
If you do not pass, you can retake the exam 24 hours after your first attempt. For further retakes the wait varies; check Microsoft's current retake policy when you schedule.
Maintaining Your Certification
Microsoft associate certifications expire annually. You keep the Azure AI Apps and Agents Developer Associate certification current by passing a free online renewal assessment on Microsoft Learn before the expiry date. There is no need to resit the full exam.
Moving Forward After AI-103
The Azure AI Apps and Agents Developer Associate certification shows you can build and operate AI apps and agents on Azure using Microsoft Foundry. If your role extends past AI implementation into wider solution design, AZ-305 covers the broader Azure architecture knowledge. For how AI-103 fits into the wider Azure AI Engineer role, see our Azure AI Engineer career path guide.
The domains on this exam, particularly agent orchestration and responsible AI controls, describe day-to-day engineering work, so keep practising against real Foundry projects after you pass.
Practise for AI-103 with scenario questions across all five domains before you sit the real exam.