Microsoft AI-103 Exam Guide: Syllabus, Cost, Passing Score and 6-Week Study Plan

The Microsoft AI-103 exam is the exam to book if you build AI applications and agents on Azure. It earns the Microsoft Certified: Azure AI Apps and Agents Developer Associate credential, and it targets working Azure AI engineers who write Python and ship solutions built on Microsoft Foundry. AI-103 replaced AI-102, the old Azure AI Engineer Associate exam, which Microsoft retired on 30 June 2026. So if your study folder still holds AI-102 notes, this guide maps what changed, what the new exam tests, and how to prepare in six weeks.

Two tools make the plan easier to follow. Our free study timetable generator turns the six-week plan below into a personal weekly schedule in minutes. And if your role leans toward solution design rather than code, our Microsoft AB-100 exam guide covers the architect-level agentic AI credential instead.

What Is the Microsoft AI-103 Exam?

The Microsoft AI-103 exam, Developing AI Apps and Agents on Azure, certifies Azure AI engineers who build, manage and deploy agents and generative AI solutions with Microsoft Foundry and Python. It replaced AI-102, tests five weighted domains, and needs a score of 700 to pass.

In plain terms, Microsoft rebuilt its Azure AI developer credential around the way teams now work. The center of the exam is Microsoft Foundry, where you choose models, ground them in your own data, build agents, and run those agents in production. Vision, language and document skills still appear, but inside a generative AI workflow rather than as standalone services to memorize. It is an Associate-level certification, above fundamentals exams and below expert credentials such as AB-100.

AI-102 vs AI-103: What Actually Changed

Plenty of AI-102 material still circulates online, and some of it remains useful. Azure AI services, responsible AI principles and Python skills all carry over. The exam itself, though, has a different center of gravity, and studying the old outline as your main plan is a genuine risk.

Point of comparison	AI-102 (retired)	AI-103 (current)
Credential earned	Azure AI Engineer Associate	Azure AI Apps and Agents Developer Associate
Status	Retired on 30 June 2026	Current exam, skills measured as of 16 April 2026
Platform focus	Azure AI services such as vision, language, search and document intelligence	Microsoft Foundry first: models, grounding, agents and production operations
What you build	Applications that call prebuilt AI services	Generative AI applications, RAG pipelines and orchestrated multi-agent systems
Heaviest topic area	Spread across planning and individual service domains	Generative AI and agentic solutions at 30-35% of the exam
Best suited to	Engineers integrating managed AI services into existing apps	Engineers who design, build and run agents and AI apps end to end

The practical takeaway is simple. Keep your AI-102 notes on core services and Python, but rebuild your study plan around Foundry, retrieval-augmented generation (RAG), agent design and evaluation. Those topics decide this exam.

Microsoft AI-103 Exam Facts at a Glance

Fact	Detail
Exam code and name	AI-103: Developing AI Apps and Agents on Azure
Credential	Microsoft Certified: Azure AI Apps and Agents Developer Associate
Level	Associate
Exam price	US $165 in the United States; the price varies by the country where the exam is proctored
Duration	Most prep guides list 120 minutes; Microsoft does not publish the duration on the certification page, so check your booking confirmation
Passing score	A score of 700 or greater on Microsoft scaled scoring
Prerequisite exams	None. Passing AI-103 alone earns the certification
Skills measured	Five domains: 25-30%, 30-35%, and 10-15% for each of the remaining three
Renewal	Associate certifications expire annually and renew through a free online assessment on Microsoft Learn

One booking detail matters: register with a personal Microsoft account, not a work or school account.

Microsoft AI-103 Exam Skills Breakdown: Where to Focus

Microsoft publishes the skills outline with weights, and the weights tell you where your study hours earn the most marks. Two domains carry 55-65% of the exam between them, so plan and manage plus generative and agentic work deserve the bulk of your timetable.

Domain	Weight	What it covers	Study focus
Plan and manage an Azure AI solution	25-30%	Choosing models and Foundry services, setting up projects, CI/CD, quotas and cost, security with managed identity and keyless credentials, monitoring, responsible AI guardrails and auditing	Learn to justify each design choice: which model, which retrieval method, which security control. Practice reading a scenario and picking the service that fits its constraints.
Implement generative AI and agentic solutions	30-35%	Building generative apps with Foundry SDKs, RAG, agent roles and tool schemas, function calling, conversation memory, multi-agent orchestration, evaluation, tracing and token analytics	This is the exam. Build at least two working agents by hand, one of them grounded in your own documents through RAG, and learn how evaluation catches weak answers.
Implement computer vision solutions	10-15%	Image and video generation and editing, multimodal understanding, captions and visual question answering, Content Understanding pipelines, visual safety and prompt injection risks in images	Focus on workflow design rather than deep vision theory: which pipeline, which control, which safety check.
Implement text analysis solutions	10-15%	Entity and summary extraction, structured JSON output, sentiment and safety detection, translation, speech to text and text to speech, retrieval and grounding pipelines	Connect every topic back to agents. Speech and retrieval appear as agent modalities and tools, not as isolated features.
Implement information extraction solutions	10-15%	Document pipelines that combine OCR, layout analysis and field extraction, Content Understanding analyzers, structured and markdown outputs for downstream reasoning	Practice the full path from a messy PDF to clean, grounded data that an agent or RAG index can trust.

A useful rule follows from the table. If a topic helps an agent act on real data safely, it matters on this exam. Standalone service features matter less than they did under AI-102.

Who Should Take AI-103, and Who Should Not

Take AI-103 if you write Python for a living or are moving into AI engineering, and your work involves Azure. It fits developers who build chat and agent experiences, engineers who ground models in company data, and AI-102 holders who need a current credential that reflects generative AI work.

Skip it for now if you do not code. The exam assumes you can read and reason about Python and application design, so a business user will struggle no matter how well they know Copilot from the user side. Business-focused readers do better with a fundamentals or business AI credential first. Skip it, too, if your stack lives entirely outside Azure: the skills are deep in Microsoft Foundry, and the credential proves Azure expertise rather than general AI knowledge.

Prerequisites and Skills to Build First

Microsoft lists no prerequisite exam for AI-103, but the audience profile sets a clear bar. Before you book, you want four foundations in place.

• Python development experience. Microsoft states plainly that candidates should have experience developing apps in Python. You need comfortable reading of functions, classes, JSON handling and API calls.
• Working knowledge of Azure services. Resource groups, identity, networking basics, storage and monitoring appear throughout the scenarios.
• Generative AI concepts. Tokens, embeddings, prompts, grounding and RAG should be familiar ideas before you open the detailed outline.
• Hands-on time in Microsoft Foundry. Deploy a model, connect a data source, and run an agent in the portal and through the SDK. Concepts stick when your hands have done the work.

None of this requires another certification first. A developer with six months of Python and a few weekends in Foundry can start the six-week plan below with confidence.

A 6-Week Microsoft AI-103 Exam Study Plan

This plan assumes about 8 to 10 study hours a week alongside a job. Compress it if you already build on Azure daily; stretch the agent weeks if Foundry is new to you. Every week ends with a build task, because this exam rewards people who have done the work. If your weekly hours are limited, split week 5 across two weeks, covering vision first and then text analysis and information extraction, rather than rushing all three smaller domains at once.

Week	Focus	Build or finish
Week 1	Foundations: Foundry tour, model catalog, Python SDK setup, tokens, prompts and embeddings	Deploy one model and call it from a Python script
Week 2	Plan and manage (25-30%): model and service selection, deployment options, CI/CD, quotas, cost, security and responsible AI controls	Secure a Foundry project with managed identity and keyless credentials
Week 3	Generative AI core: RAG design, indexing, retrieval methods, grounding quality and evaluation	Build a RAG app over your own documents and test its answers
Week 4	Agents (the heart of the 30-35% domain): roles, tools, function calling, memory, multi-agent orchestration, tracing and safeguards	Build a two-agent workflow with an approval step and full tracing
Week 5	Vision, text and information extraction (10-15% each): multimodal workflows, speech, translation, OCR and Content Understanding pipelines	Run one document extraction pipeline from PDF to structured JSON
Week 6	Revision and rehearsal: official practice assessment, exam sandbox, weak-domain drills, timed question sets	Two timed practice runs, then book the exam while the material is fresh

Treat the tracker that comes with this guide as your control sheet. Tick off each domain topic as you cover it, and let the empty boxes in week 6 tell you exactly what to revise.

Free Official Resources That Actually Help

Microsoft publishes enough free material to prepare well without buying a course on day one. Use these sources in this order.

• The official AI-103 study guide on Microsoft Learn lists every skill the exam measures, with the weights used in the table above. Make it your syllabus and check topics off against it.
• The certification page for Azure AI Apps and Agents Developer Associate links the official self-paced learning path and the exam sandbox, where you can explore the real exam interface before test day.
• The practice assessment now runs through AI Skills Navigator, and Microsoft notes that you must sign in there to launch it. It gives the closest free preview of question style.
• Microsoft Learn training modules cover Foundry, RAG and agent development step by step, with sandbox exercises that cost nothing.

One honest warning: some third-party question banks still teach the retired AI-102 outline. Check any paid resource against the official study guide before you trust it.

Booking, Retakes and Renewal

You schedule AI-103 through your Microsoft Learn certification profile, and the price shown depends on the country where the exam is proctored: US $165 in the United States, with different figures elsewhere. You book the exam through Pearson VUE and can take it at a test center or online with a proctor. Microsoft runs one retake policy across its role-based exams, so read the current policy on the exam page before you book; the standard rule lets you retake 24 hours after a first failed attempt, with longer waits after later attempts.

After you pass, one task remains. Microsoft associate certifications expire every year, and you renew this one by passing a free online assessment on Microsoft Learn inside the renewal window. Set a reminder for month ten and stay ahead of it.

Common Mistakes That Cost Candidates the Pass

• Studying the AI-102 outline. The retired exam weighted individual services far more heavily. AI-103 rewards Foundry, agents and RAG, so an old plan misallocates most of your hours.
• Reading without building. Scenario questions describe real configurations and failure modes. Candidates who have deployed agents and fixed grounding problems recognize the right answers faster.
• Ignoring the plan and manage domain. At 25-30%, security, cost, monitoring and responsible AI controls carry nearly a third of the exam. Pure builders lose easy marks here.
• Skipping evaluation. Fabrications, relevance checks, safety evaluations and tracing run through the outline. Treat evaluation as a core skill, not a footnote.
• Booking with a work account. Microsoft warns that records tied to an employer account are lost when you leave. Register with a personal account from the start.

One-Page Revision Box

• AI-103 = Developing AI Apps and Agents on Azure; credential: Azure AI Apps and Agents Developer Associate.
• It replaced AI-102, retired 30 June 2026. New candidates cannot book AI-102.
• Biggest domain: generative AI and agentic solutions, 30-35%. Second: plan and manage, 25-30%.
• Vision, text analysis and information extraction carry 10-15% each.
• Core stack: Microsoft Foundry, Python, RAG, function calling, multi-agent orchestration, tracing.
• Pass mark: a score of 700 or greater. Price: US $165 in the United States, varies by country.
• Renew yearly through a free Microsoft Learn assessment.

Frequently Asked Questions

Is AI-103 a direct replacement for AI-102?

Yes. Microsoft retired AI-102, the Azure AI Engineer Associate exam, on 30 June 2026, and AI-103 is the current exam for engineers who build AI apps and agents on Azure. Existing AI-102 holders keep their credential until its expiry date, but new candidates must take AI-103.

Do I need to pass any exam before AI-103?

No. Microsoft lists no prerequisite exam for this certification. You do need real Python development experience and familiarity with Azure services, because the questions assume both from the first section onward.

How much Python does AI-103 require?

Enough to build with the Foundry SDKs confidently: calling APIs, handling JSON, structuring functions and reading unfamiliar code under time pressure. You write and reason about application code, so a scripting-only background needs topping up before the exam.

What score do I need to pass AI-103?

Microsoft requires a score of 700 or greater. The score uses Microsoft scaled scoring, so treat 700 as the official pass mark rather than a simple percentage of questions answered correctly.

How much does the AI-103 exam cost?

The price is US $165 in the United States. Microsoft sets the fee by the country where the exam is proctored, so check the certification page for your local price before you schedule.

How long should I prepare for AI-103?

Most working developers need five to six weeks at 8 to 10 hours a week, weighted toward hands-on agent and RAG builds. If Foundry is entirely new to you, add a week at the start for the foundations work in week 1 of the plan above.

Your Next Step

Download the free printable tracker that matches this guide: AI-103 2-page study tracker (PDF). Page 1 carries the exam facts and the six-week checklist; page 2 lists every domain topic from the official skills outline, ready to tick off as you master it. Print it, pin it above your desk, and let it drive your revision.

Preparing with a friend or a study group? Share this guide on WhatsApp or Telegram so you can hold each other to the six-week plan.