How to Pass the Microsoft AI-103 Exam in 2026: An 8-Week Study Plan for the New Azure AI App and Agent Developer Certification

If you have been eyeing the Azure AI Engineer certification, the ground has just shifted under your feet. Microsoft is retiring the AI-102 exam on 30 June 2026, and the brand new AI-103 exam, Developing AI Apps and Agents on Azure, takes its place. This guide gives you a realistic 8-week AI-103 study plan, breaks down every exam domain, and shows you exactly what changed so you do not waste a single hour studying the wrong material.

The short answer to "can I pass AI-103 in eight weeks?" is yes, if you already write Python and have touched Azure before. AI-103 leads to the new Microsoft Certified: Azure AI Apps and Agents Developer Associate credential, and it is far more focused on building agents, retrieval-augmented generation (RAG) pipelines and generative AI apps than the exam it replaces. Treat it as a developer exam, not a multiple-choice trivia test, and plan your weeks around hands-on practice.

What Is the Microsoft AI-103 Exam?

AI-103 is Microsoft's new associate-level exam for developers who build AI applications and autonomous agents on Azure using Microsoft Foundry. Passing it earns the Microsoft Certified: Azure AI Apps and Agents Developer Associate certification, the direct successor to the Azure AI Engineer Associate certification that AI-102 awarded.

Microsoft describes the ideal candidate as an Azure AI engineer who builds, manages and deploys agents and AI solutions that take advantage of Microsoft Foundry. You are expected to develop apps using Python and to be familiar with general AI, generative AI and core Azure services. In practice you collaborate with solution architects, data scientists, DevOps engineers and security engineers to ship production AI systems.

The headline shift is agents. Where AI-102 was built around wiring up individual Azure AI services, AI-103 is built around generative AI, agent orchestration, RAG and multimodal apps. If you have been following the move from "call the vision API" to "build an agent that reasons, retrieves and acts," AI-103 is the exam that finally reflects it.

Exam Tip: AI-103 assumes you can read and write Python. If you cannot comfortably call an SDK, parse a JSON response and loop through results in code, spend your first week on Python fundamentals before anything else. This is a developer certification, and the questions reward people who have actually built something.

AI-103 Exam Format at a Glance

Microsoft publishes the official skills and the passing score, but holds back the exact question count, which varies between sittings. The figures below combine the official AI-103 study guide with the typical format of a current Microsoft associate exam, so treat the question count and timing as a realistic range rather than a fixed promise.

The 700 out of 1000 pass mark is a scaled score, not a simple percentage, so do not assume you need exactly 70 percent of questions correct. Aim for 80 percent or higher on practice exams before you book, which gives you a comfortable buffer on exam day.

AI-103 vs AI-102: What Actually Changed?

If you started studying for AI-102, the good news is that a lot of your knowledge still counts. The bad news is that the centre of gravity has moved. Here is how the two exams compare.

The single biggest change is Microsoft Foundry. AI-103 expects you to know Foundry projects, the Foundry Agent Service, the Microsoft Agent Framework and the modern Azure AI SDKs. If your notes still say "Azure AI Studio," they are out of date. Spend time in the current Foundry portal so the terminology in the questions matches what you have actually clicked through.

For a wider view of how this sits against other AI credentials, see our guide on whether AI certifications actually get you hired and our comparison of the AWS AI Practitioner and Azure AI Fundamentals.

The Five AI-103 Exam Domains and Their Weights

Microsoft groups the AI-103 skills into five domains. The percentages tell you exactly where to spend your study time, so weight your effort accordingly. The two generative and planning domains together make up well over half the exam.

1. Plan and manage an Azure AI solution (25 to 30 percent). Choosing the right Foundry models and services, designing infrastructure, configuring deployments and CI/CD, managing quotas and cost, securing systems with managed identity and private networking, and applying responsible AI controls.
2. Implement generative AI and agentic solutions (30 to 35 percent). The heart of the exam. Building generative apps with Foundry, implementing RAG, defining agent roles and tool schemas, function calling, conversation memory, multi-agent orchestration, prompt engineering and observability.
3. Implement computer vision solutions (10 to 15 percent). Image and video generation from prompts, inpainting and mask-based edits, multimodal understanding, captioning, accessibility alt-text and Content Understanding pipelines.
4. Implement text analysis solutions (10 to 15 percent). Entity, topic and sentiment extraction, structured JSON output, translation with Azure Translator, and speech-to-text and text-to-speech for agent interactions.
5. Implement information extraction solutions (10 to 15 percent). Ingesting and indexing content, semantic and vector search, RAG ingestion with OCR, and document extraction with Content Understanding.

Exam Tip: Domains two and one are worth up to 65 percent of your marks combined. If you are short on time, master generative AI, agents and solution planning first. The three smaller domains share the remaining marks, so a thin pass there still leaves you well above the 700 line if your agent knowledge is solid.

Your 8-Week AI-103 Study Plan

This plan assumes around 8 to 10 hours of study a week, mixing Microsoft Learn modules with hands-on building in your own Azure subscription. Adjust the pace if you already work with Azure AI daily. The golden rule: never study a domain without building something in it the same week.

Weeks 1 to 2: Foundations and Foundry Setup

• Confirm your Python is solid: SDK calls, JSON parsing, async basics and error handling.
• Set up an Azure subscription and create your first Microsoft Foundry project.
• Work through the official AI-103 study guide and the Develop AI apps and agents on Azure learning path.
• Deploy your first model in Foundry and call it from a short Python script.
• Learn the vocabulary: Foundry projects, Foundry Tools, the Agent Service and the Agent Framework.

Weeks 3 to 4: Generative AI and Agents (the big domain)

• Build a basic generative app that calls an LLM and handles the response safely.
• Implement a simple RAG pipeline: ingest documents, index them, retrieve and ground answers.
• Build your first agent with a clear role, a tool schema and conversation memory.
• Add function calling so your agent can hit an external API.
• Experiment with multi-agent orchestration and an approval flow for a risky action.
• Practise prompt engineering and tweak model parameters to see how output changes.

Weeks 5 to 6: Vision, Text and Information Extraction

• Generate images from prompts and try inpainting and mask-based edits.
• Build a multimodal solution that captions images and answers questions about them.
• Extract entities, sentiment and structured JSON from text using generative prompting.
• Add speech-to-text and text-to-speech to an agent so it can hold a voice interaction.
• Build a retrieval pipeline with semantic and vector search, then extract fields from a PDF using Content Understanding and OCR.

Week 7: Responsible AI, Security and Operations

• Configure safety filters, content moderation and guardrails on a generative app.
• Set up monitoring: tracing, token analytics, latency breakdowns and drift detection.
• Secure a solution with managed identity, keyless credentials and private networking.
• Implement evaluators that detect fabrications and check grounding quality.
• Review cost management: quotas, scaling and rate limits.

Week 8: Practice Exams and Targeted Revision

• Sit full-length AI-103 practice exams under timed conditions.
• Log every wrong answer and map it back to its domain.
• Re-study only your weak domains, do not waste time on what you already know.
• Re-read the official study guide one final time to catch anything Preview-related.
• Book the exam for the end of the week while the material is fresh.

Exam Tip: Microsoft notes that most questions cover generally available (GA) features, but commonly used Preview features can appear. Do not panic if you see an unfamiliar Foundry feature. Reason from the underlying concept, agents, grounding or safety, rather than memorising every menu name.

Common AI-103 Mistakes to Avoid

Even strong developers trip over the same things. Sidestep these and you remove most of the risk.

• Studying AI-102 material. Plenty of older guides and dumps still describe Azure AI Studio and the old service-by-service approach. They will actively mislead you. Anchor on the current AI-103 study guide dated 16 April 2026 or later.
• Reading instead of building. This is a developer exam. You cannot pass on theory alone. If you have not built an agent, a RAG pipeline and a generative app yourself, you are not ready.
• Ignoring responsible AI. Safety filters, evaluators, guardrails and content moderation are woven through multiple domains, not bolted on at the end. Treat them as core, not optional.
• Underweighting the planning domain. Choosing the right model and Foundry service for a task is 25 to 30 percent of the exam. Know when to reach for an LLM, a small model, a multimodal model or vector search.
• Cramming the smaller domains. Vision, text and information extraction each carry 10 to 15 percent. A little hands-on practice in each goes a long way and protects your overall score.

Is AI-103 Worth It in 2026?

For developers moving into AI engineering, yes. AI-103 sits exactly where the market is heading: building agents and generative apps rather than calling standalone APIs. Azure professionals with combined AI and cloud skills consistently command a salary premium over non-certified peers, and "build me an agent" is fast becoming a core enterprise ask.

Because AI-103 is brand new, there is also a timing advantage. Far fewer people hold it than held AI-102, so the credential signals that you are current with Foundry, agents and RAG rather than a generation behind. If you were going to take AI-102, there is no reason to chase a retiring exam. Put your eight weeks into AI-103 instead.

Ready to Start Practising?

Reading about agents and RAG will only take you so far. The candidates who pass AI-103 first time are the ones who drilled realistic questions until the exam format held no surprises, then walked in knowing exactly how the scenarios are framed.

CertCrush gives you exam-style practice questions with full explanations, so you can find your weak domains fast and fix them before exam day. Browse our certification courses to see how we structure practice for Azure and AI exams, and check our pricing to find the plan that fits your timeline.

Create your free CertCrush account and start practising for AI-103 today. Eight focused weeks, a clear plan and enough hands-on reps, and that Azure AI Apps and Agents Developer Associate badge is well within reach.