How long to study for AI-103? 3–16 weeks, by background

Short answer, and these are our estimates, not official figures: if you're an Azure developer new to generative AI, plan on about 4–6 weeks of part-time study at roughly 6–8 hours a week. If you code in Python but are new to Azure, plan on about 8–10 weeks. If you're changing careers and still learning to code, expect 12–16 weeks or more. Either way, a real share of that time should be hands-on in Microsoft Foundry, not just reading.

AI-103 is Microsoft's associate-level exam for developers building AI apps and agents on Azure: five domains, 120 minutes, and scenarios where you pick the service, setting, role or agent tool that fits a constraint. Microsoft expects Python app experience and familiarity with generative AI and Azure, so your timeline depends mostly on which of those three you already have. Here's the timeline by background, how to split your hours across the domains, and how to tell when you're ready.

Study time by background

These are rough ranges based on what the outline covers, not guarantees, and not numbers Microsoft publishes. Here's what each background brings:

• Python developer new to Azure. You can read the SDK code in the questions, but Azure's resource model, Microsoft Entra ID, RBAC roles, private networking and quotas are new. Planning and managing is 25–30% of the exam and is mostly Azure knowledge, so budget extra time there.
• Azure developer new to generative AI. Identity, networking and deployments are familiar; models, prompts, RAG, agents, tools, guardrails and evaluators are new. Most of your time goes into the generative AI and agents domain, the largest at 30–35%.
• AI-102 holder. Many concepts carry over, but much of the platform has been renamed or rebuilt: Microsoft Foundry projects instead of hubs, Foundry Agent Service with conversations and responses instead of threads and runs, Guardrails instead of content filters, and gpt-image instead of DALL-E 3. Your time goes into unlearning old answers.
• Career changer. Microsoft expects experience developing apps in Python. Learn enough Python to write and debug a small script that calls an API first, then start AI-103. If AI concepts are new too, Microsoft's AI-901 (Azure AI Fundamentals) is an optional first step. Our is AI-103 worth it? post covers who the cert fits.

The exam at a glance

What drives the time: the domain weights

Weight your time the way Microsoft weights the exam. The right-hand column is our arithmetic: the official weight ranges applied to a 40–60 hour plan, which is roughly what 6–8 weeks at 6–8 hours a week works out to. On a shorter or longer track, keep the same proportions.

• Plan and manage is the Azure-heavy domain: choosing models and Foundry Tools, the Foundry resource and project model, deployment types (Global Standard, Data Zone, Provisioned, Batch), quotas and HTTP 429 handling, tracing to Application Insights, keyless auth with Microsoft Entra ID, RBAC roles such as Foundry User (formerly Azure AI User), private networking, and responsible AI with Guardrails, Prompt Shields and evaluators.
• Generative AI and agents is the biggest: connecting an app with the Foundry SDK and the Responses API, RAG, structured outputs, model parameters and reasoning models, Foundry Agent Service (prompt and hosted agents, conversations, tools such as File Search, Azure AI Search, Code Interpreter, MCP and A2A), tool approval, memory, and multi-agent patterns with Microsoft Agent Framework.
• Computer vision covers image generation and editing with the gpt-image models, video generation (Sora 2, preview), vision-capable chat models for captions and visual Q&A, Content Understanding for images and video, and responsible AI for multimodal content.
• Text analysis covers using language models for analysis tasks, Azure Language in Foundry Tools (PII, entities, language detection), Azure Translator, and Azure Speech, including Voice Live for voice agents.
• Information extraction covers Azure AI Search ingestion, hybrid and vector queries, the semantic ranker, skills and integrated vectorization, agentic retrieval and Foundry IQ, and document extraction with Content Understanding and Document Intelligence.

Two timeline profiles

Profile A — the 6-week track (Azure developer, AI-102 holder, or already using Foundry)

• Week 1 — plan and set up: take a mixed practice set on day 1 to find your gaps, then models and Foundry Tools, the Foundry resource and project model, deployment types, endpoints and RBAC.
• Week 2 — manage, secure and responsible AI: quotas, monitoring and tracing, private networking, Guardrails, Prompt Shields and evaluators.
• Week 3 — generative AI apps: the Foundry SDK and Responses API, RAG, structured outputs, parameters and reasoning models.
• Week 4 — agents: Foundry Agent Service, tools, MCP approval, memory, multi-agent orchestration and tracing.
• Week 5 — vision, text and extraction, then a mixed quiz across all five domains.
• Week 6 — mocks: Microsoft's free Practice Assessment plus full timed mocks; drill what's under 80%, then book.

Profile B — the 10-week track (Python developer new to Azure)

• Weeks 1–3 — plan and manage. Set up a subscription and your first Foundry project in week 1; this is where you learn Azure's identity, roles and networking, which everything else builds on.
• Weeks 4–7 — generative AI and agents. Apps and RAG first, then single agents and tools, then multi-agent patterns and operations.
• Week 8 — computer vision.
• Week 9 — text analysis and information extraction.
• Week 10 — review and mocks: re-drill everything under 80%, take the Practice Assessment and full timed mocks, then sit the exam.

Want both tracks written out week by week, with a daily routine, four cheat sheets and the Azure CLI commands to know? That's our AI-103 study plan.

How to use your practice-question scores

• Track accuracy per domain, not overall. 82% overall with generative AI and agents at 65% isn't ready; that domain is 30–35% of the exam.
• Only count fresh questions. A second pass over questions you've already seen measures memory. Readiness is consistently 80% or higher on questions you haven't seen.
• Read the explanation for every miss, including why each wrong answer is wrong. On AI-103 the wrong answers are usually real services or settings that would be right for a slightly different scenario.
• Use Microsoft's free Practice Assessment on AI Skills Navigator (aka.ms/AI103-PA) as an official check on question style.
• Finish with a full timed mock at 80%+. CrushCert's mock is 50 questions on a 120-minute timer with a 70% pass line; treat 70% as the floor, not the target.

What makes it take longer (or shorter)

• Skipping hands-on time. Reading about agent tool approval is not the same as watching your agent stop and wait for an approval response. People who only read tend to need more weeks, not fewer.
• Mixing up the lookalikes. File Search vs the Azure AI Search tool vs Foundry IQ, JSON mode vs structured outputs, Global Batch vs Provisioned, Groundedness vs Relevance evaluators, Foundry User vs Contributor, Document Intelligence vs Content Understanding. Drill them side by side.
• Studying from outdated material. Tutorials from before the renames will teach Azure AI Studio, hubs, threads and runs, content filters, Semantic Kernel and DALL-E 3. Use current Microsoft Learn pages and a bank written for the April 16, 2026 outline.
• Already building in Foundry? Take a full practice set on day one and spend your time on the domains you don't use at work, often vision, speech and document extraction. That's how the 3-week end of the range happens.

How to study efficiently

• Pair free training with a question bank. Microsoft's free self-paced AI-103 training and exam prep videos cover the material; practice questions show you where you're actually weak.
• Use adaptive quizzes so your reps go to the topics you miss rather than the ones you already know.
• Type the setup commands once. CrushCert's Azure CLI lab has you create a Foundry resource, a project, a Global Standard model deployment and a Foundry User role assignment with az cognitiveservices and az role assignment commands. It simulates the commands (nothing is deployed), so use it for syntax and a real subscription for building.
• Pick your question sources deliberately. Our best AI-103 practice tests post compares the free and paid options.

Is it worth the time?

For developers building AI apps and agents on Azure, usually yes: it's Microsoft's current role-based credential for that work, and it renews yearly for free. Non-coding roles and people on other clouds have better-fitting options. The full take is in is AI-103 worth it?, and the AI-103 practice test page shows exactly what's in the prep.

Keep reading

Frequently asked questions