Last updated: 22 October 2026. Exam details, prices and every course below were read on Microsoft Learn and Udemy on that date. Written by Josh Hutcheson, OnlineCourseing editor. See our review methodology.
Some links on this page are affiliate links: if you buy through them we may earn a commission at no extra cost to you. It never changes the verdict, and we name options we earn nothing from when they are the better fit.
AI-103 is Microsoft’s exam for building AI apps and agents on Azure. It earns the Microsoft Certified: Azure AI Apps and Agents Developer Associate, and since Microsoft retired AI-102 in 2026 it is the associate-level certification for developers who build generative AI applications on Azure. If you were planning to sit AI-102, this is the exam you take now.
This guide covers what AI-103 tests and how the five skill areas are weighted, what changed from AI-102, how it differs from AI-200, what the exam costs, how hard it is and how long to study, and which free and paid resources are worth your time. Every fact was checked against Microsoft’s own credential page and study guide on 22 October 2026.
60-SECOND VERDICT
Bottom line: Start with Microsoft Learn’s free 30-hour AI-103 course and do the labs in your own Azure subscription. Add one current Udemy exam-prep course if you want structure, and take Microsoft’s free practice assessment before you book.
- Exam: AI-103, Developing AI Apps and Agents on Azure; $165 in the US; 700 to pass
- Who it is for: Python developers who build generative AI apps or agents on Azure
- Best free prep: Microsoft Learn: four learning paths, about 30 hours, plus a free practice assessment
- Best paid course: Kuljot Singh Bakshi’s AI-103 Exam Prep on Udemy, updated October 2026
- Best for a portfolio: Udacity’s Microsoft Agentic AI Nanodegree: four reviewed agent projects
- Study time: Four to six weeks at 6 to 8 hours a week for a working developer
What is the AI-103 certification?
Before you spend money on the wrong online course, read this.
Get the free 2026 Platform Comparison Guide — 12 platforms compared on price, certificates, and refund policies. Instant PDF, plus my honest Tuesday picks.
No spam. Unsubscribe anytime.
According to Microsoft’s credential page, the Azure AI Apps and Agents Developer Associate “validates your expertise in designing, developing, and deploying advanced Azure AI solutions using Python and Microsoft Foundry.” Microsoft describes the candidate as an Azure AI engineer who builds, manages and deploys agents and AI solutions on Foundry, with experience developing apps in Python and familiarity with generative AI and Azure services.
In practice that means five kinds of work, which map directly to the exam’s skill areas:
- Planning and managing an Azure AI solution: choosing models and Foundry services, deploying them, monitoring cost and quality, and securing them.
- Building generative AI apps and agents: retrieval-augmented generation (RAG), tool use, multi-agent workflows and evaluation.
- Computer vision: image and video generation, and multimodal understanding of images and video.
- Text analysis: entity and summary extraction with language models, translation and speech.
- Information extraction: ingesting and indexing content for search and grounding, and pulling fields out of documents.
The level is intermediate. It is a role-based associate certification, a step above AZ-900 and Microsoft’s AI fundamentals exam, and it assumes you can already write Python against an API.
AI-103 exam at a glance
| Detail | AI-103 (October 2026) |
|---|---|
| Full name | Exam AI-103: Developing AI Apps and Agents on Azure |
| Certification | Microsoft Certified: Azure AI Apps and Agents Developer Associate |
| Price | $165 in the US; set by the country where the exam is proctored |
| Time | 120 minutes listed; Microsoft’s duration table gives associate exams without labs 100 minutes of exam time inside a 120-minute seat |
| Passing score | 700 or higher (a scaled score, not a percentage) |
| Questions | Not published; one July 2026 candidate reported 63, including a 7-question case study |
| Question types | Multiple choice, drag and drop, yes/no and case studies |
| Open book? | Microsoft Learn can be opened during the exam |
| Where | Pearson VUE test center or online proctored |
| Languages | English plus nine others, including Spanish, French, German, Japanese and Portuguese (Brazil) |
| Prerequisites | None required; Python and Azure experience expected |
| Renewal | Every year, free, by online assessment |
| Retakes | Wait 24 hours after a first fail, then 14 days; at most five attempts in 12 months |
Two details on that table are worth knowing before you book. First, the exam allows access to Microsoft Learn, which Microsoft confirms on its exam duration and experience page for associate and expert exams. Second, if your language is not offered, you can request an extra 30 minutes. Microsoft updates the English version first, and localized versions follow about eight weeks later, so take it in English if you can.
Microsoft also recommends registering with a personal Microsoft account rather than a work or school account. If you register with your employer’s account and later leave, your exam records go with it.
AI-103 vs AI-102: what changed
AI-102, the exam for the Azure AI Engineer Associate, is gone. Microsoft’s AI-102 credential page now reads: “This certification and the renewal assessment are retired.” The exam retired on 30 June 2026, and AI-103 is its successor.
The two exams share a lot of ground: both cover vision, language, search and document extraction on Azure. The difference is where the weight sits.
| AI-102 (retired) | AI-103 (current) | |
|---|---|---|
| Certification | Azure AI Engineer Associate | Azure AI Apps and Agents Developer Associate |
| Center of gravity | Azure AI services (Vision, Language, Speech, Document Intelligence) | Microsoft Foundry: generative AI apps and agents |
| Agents | One section among several | The largest domain, with generative AI (30–35%) |
| Responsible AI | Content safety basics | Guardrails, safety evaluations, tracing, approval workflows and agent oversight |
| Vision | Image analysis and custom vision | Image and video generation and multimodal understanding |
| Status | Retired; renewal assessment retired | Current; skills measured as of 16 April 2026 |
If you already hold AI-102, it stays valid until its expiry date, but you cannot renew it. If you studied for AI-102 and never sat it, much of what you learned about Azure AI Search, Language and Document Intelligence still applies. The gap is Foundry: building and deploying agents, orchestrating multiple agents, evaluating them and monitoring them in production.
WATCH OUT
Many courses and question banks still carry AI-102 material under an AI-103 title. Before you buy, check the last-updated date and look for Foundry agents, the Agent Service and multi-agent orchestration in the curriculum. Coursera, for example, still lists an AI-102 program for the retired exam and had no AI-103 course when we searched on 22 October 2026.
AI-103 vs AI-200: which Azure exam should you take?
Microsoft split the old developer track into two exams in 2026. AI-200, the Azure AI Cloud Developer Associate, replaced AZ-204 and covers the back end behind AI applications: container hosting, Azure Container Apps, Kubernetes, Cosmos DB and data pipelines. AI-103 covers the AI layer on top: models, agents, retrieval and safety.
| If your work is mostly… | Take |
|---|---|
| Building chat apps, copilots, RAG systems or agents on Foundry | AI-103 |
| Building the APIs, containers, databases and pipelines those apps run on | AI-200 |
| Both, on a small team | AI-103 first if the AI features are the product; AI-200 first if you run the platform |
| Administering Azure infrastructure | Neither yet; start with AZ-104 |
Both exams cost $165, take 120 minutes and need 700 to pass, so the choice is about the job you do, not difficulty or price.
AI-103 skills measured: the five domains
Microsoft’s AI-103 study guide lists the skills measured as of 16 April 2026. The weights tell you where to spend your time: the first two domains are more than half the exam.
| Domain | Weight | What it tests |
|---|---|---|
| Plan and manage an Azure AI solution | 25–30% | Choosing Foundry services and models (large, small, multimodal), deployment options, CI/CD, quotas and cost, monitoring drift and grounding quality, security (managed identity, private networking, keyless credentials) and responsible AI controls |
| Implement generative AI and agentic solutions | 30–35% | RAG, multistep and tool-augmented workflows, evaluation, Foundry SDKs, building agents with retrieval, function calling and memory, multi-agent orchestration, approval flows, tracing and tuning |
| Implement computer vision solutions | 10–15% | Image and video generation and editing, captions and alt text, visual question answering, Content Understanding for images and video, and filters for unsafe visual content |
| Implement text analysis solutions | 10–15% | Entity, topic and summary extraction with language models, sentiment and safety detection, translation, and speech to text and text to speech for agents |
| Implement information extraction solutions | 10–15% | Ingesting and indexing documents, images, audio and video; semantic, hybrid and vector search; OCR and layout extraction with Content Understanding |
Three themes run through all five domains and show up again and again in candidate reports:
- Security by identity, not keys. The guide asks for managed identity and keyless credentials, and candidates report that questions assume Microsoft Entra ID authentication rather than API keys.
- Responsible AI as engineering. Content filters, Prompt Shields against prompt injection, safety evaluations and audit trails appear in the planning domain and again in vision.
- Reading code. You are expected to read Python that calls Foundry and Azure OpenAI, including structured outputs defined with JSON schema.
Microsoft also notes that the exam may include features that are still in preview if they are commonly used, so study the current Foundry documentation rather than an older book.
How hard is AI-103?
AI-103 is an intermediate exam, and for most candidates it is moderately hard. How hard it feels depends almost entirely on how much you have built on Azure. A developer who has deployed a RAG app or an agent in Foundry will recognize most scenarios. Someone who has only watched videos will find the code and configuration questions difficult, because they ask which option works, not what a service is called.
A candidate who passed in July 2026 wrote a detailed account of the exam. Four points from it are useful for anyone preparing:
- The case study came last. The exam had 63 questions in 100 minutes, with one case study of 7 questions at the end rather than the start. The candidate advises budgeting 20 to 25 minutes for it.
- Looking things up costs time. Microsoft Learn is available, but pages load slowly and one lookup becomes several. Flag uncertain questions and come back.
- Newer topics carry weight. The candidate saw several questions on Content Understanding in Foundry and on Prompt Shields, areas that older study material covers thinly.
- Know the SDK. Questions asked about the Responses API, the Python SDK and JSON schema for structured outputs and tool definitions.
That is one person’s exam; Microsoft varies the questions, and you may see a different number or more than one case study. The pattern matches the study guide, though: AI-103 rewards people who have typed the code.
What you should know before you start
- Python: functions, classes, async calls and working with JSON.
- Azure basics: subscriptions, resource groups, role-based access and managed identities. AZ-900 covers this if you are new.
- Generative AI basics: prompts, tokens, embeddings and what RAG does.
- An Azure subscription for labs. A free account is enough to start; set a budget alert before you deploy models.
How long does it take to study for AI-103?
Microsoft’s free course for the exam, Develop AI apps and agents on Azure, is four learning paths: generative AI apps (about 7 hours), AI agents (about 10 hours), natural language solutions (about 6 hours) and visual data (about 7 hours). That is roughly 30 hours of reading and labs.
Most candidates need more than the course alone. Plan on the course plus time for your own lab work and practice tests:
| Your background | Typical preparation | At 6–8 hours a week |
|---|---|---|
| Already building AI apps on Azure | 25–35 hours | 3–4 weeks |
| Python developer, new to Azure AI | 40–60 hours | 6–8 weeks |
| Held or studied for AI-102 | 25–40 hours, focused on Foundry agents | 4–5 weeks |
| New to Python or to Azure | 70+ hours, including fundamentals | 10+ weeks |
These are planning estimates, not Microsoft figures. Adjust them after your first practice assessment: a low score in one domain tells you where the extra hours should go.
A six-week AI-103 study plan
| Week | Focus | Output |
|---|---|---|
| 1 | Microsoft Learn: generative AI apps path; set up a Foundry project | A deployed model and a working chat app in Python |
| 2 | AI agents path (the largest); agent tools and memory | An agent that calls a function and a search index |
| 3 | Multi-agent orchestration, evaluation and tracing; responsible AI controls | An evaluated agent with content filters and tracing on |
| 4 | Natural language and visual data paths; Content Understanding | One text and one document-extraction pipeline |
| 5 | Microsoft’s free practice assessment; review every miss; first paid practice test | A list of weak sub-skills |
| 6 | Fix weak areas; second practice test; exam sandbox; book the exam | Consistent scores of 80% or more on practice tests |
Best AI-103 prep courses (free and paid)
You can pass AI-103 on free material alone if you are disciplined about labs. Paid courses help with structure and with explaining the newer agent topics. We checked every course below on 22 October 2026, including its last-updated date, because outdated AI-102 material is the most common problem.
| Option | Type | Price | Last updated | Best for |
|---|---|---|---|---|
| Microsoft Learn: Develop AI apps and agents on Azure | Free course, 4 paths, ~30 hours | Free | Current | Everyone; the official syllabus |
| Kuljot Singh Bakshi, AI-103 Exam Prep (Udemy) | 18-hour course + 2 practice tests | $19.99 on 22 Oct (Udemy prices change often) | October 2026 | A structured course built around Foundry agents |
| Sandeep Soni, AI-103: Agents, RAG & MCP (Udemy) | 21.5-hour course | Varies | October 2026 | Deeper coverage of search, RAG and agent workflows |
| Ankit Mistry, AI-103 Exam Prep (Udemy) | 21-hour course + 2 practice tests | Varies | September 2026 | Vision and language services alongside Foundry |
| Scott Duffy, AI-103 Practice Tests (Udemy) | 4 tests, 100 questions | $59.99 list, often less | September 2026 | Extra practice questions |
| Udacity Microsoft Agentic AI Nanodegree | 56 hours, 4 reviewed projects | $125 a month during the current offer | September 2026 | A portfolio of graded agent projects |
Microsoft Learn: the official free course
Start here regardless of what else you use. Microsoft’s AI-103 course has four learning paths with hands-on exercises in Foundry, it tracks the current skills outline, and it costs nothing. Its weakness is that it teaches each service rather than the exam: it will not tell you which topics carry the most weight or drill you on scenario questions. Pair it with the study guide’s domain list and a practice test.
Kuljot Singh Bakshi’s AI-103 Exam Prep: best paid course
Kuljot Singh Bakshi’s course is the Udemy option that most closely follows the current exam. Its sections cover generative AI development in Foundry, low-code agent workflows, multi-agent solutions with the Microsoft Agent Framework, agent-to-agent communication, observability and tracing, and fine-tuning. On 22 October 2026 it had a 4.5 rating from 572 reviews and 6,783 students, about 18 hours of video and two practice tests, and it was last updated in October 2026. It cost $19.99 that day; Udemy prices change often, so check before you buy.
It was originally an AI-102 course and has been rebuilt for AI-103, which is why its reviews predate the new exam. The current curriculum is agent-first, so that history matters less than the update date.
See the AI-103 Exam Prep Course →
Sandeep Soni’s AI-103: Agents, RAG and MCP
If you want more depth on retrieval, Sandeep Soni’s course spends more time on Azure AI Search, RAG, function calling, agents and agent workflows, with a separate section on the Language service. It is longer, at about 21.5 hours, rated 4.6 from 138 reviews with about 1,400 students, and was updated in October 2026. It is a good second course for the information extraction domain.
Ankit Mistry’s AI-103 Exam Prep
Ankit Mistry’s course has the most students of the AI-103 courses we checked (nearly 15,000, rated 4.4 from 1,473 reviews), largely because it began as an AI-102 course. It now pairs new Foundry sections, covering playgrounds, RAG, fine-tuning, agents and prompt flow, with its older sections on Azure AI Vision and language services. That makes it useful for the vision and text domains. It was last updated in September 2026.
Udacity Microsoft Agentic AI Nanodegree: for a portfolio
The Microsoft Agentic AI Nanodegree is not an exam-prep course, and it does not follow the AI-103 outline. What it offers is four business projects built on Microsoft Foundry, the Agent Service, Semantic Kernel and Cosmos DB, each reviewed by a human. Udacity estimates 56 hours. It suits a developer who wants evidence of agent work to show an employer, which a certificate alone does not provide, and it covers the largest AI-103 domain in depth. During Udacity’s current offer it costs $125 a month. Our full review of the Nanodegree explains what each project involves, and Udacity’s Azure Generative AI Engineer program covers RAG and model deployment if you want a second option.
See the Microsoft Agentic AI Nanodegree →
AI-103 practice tests
Practice questions are where you find out whether you can apply what you studied under time pressure. Use them in this order:
- Microsoft’s free practice assessment. Microsoft offers a practice assessment for AI-103 on its AI Skills Navigator site; you need to sign in to launch it. Take it once early to find your weak domains and again before you book.
- The exam sandbox. Microsoft’s sandbox lets you click through the question formats, including drag and drop and case studies, so the interface is not a surprise.
- A paid practice set. Scott Duffy’s AI-103 practice tests on Udemy have four tests and 100 questions, a 4.5 rating from 1,146 reviews, and were updated in September 2026. Kuljot Singh Bakshi’s course also includes two practice tests.
Aim for steady scores of 80% or more on practice tests before you book. That target is a rule of thumb, not a Microsoft figure: the real exam reports a scaled score where 700 passes, and question banks differ in difficulty. What matters more is that no domain lags, because the agent and planning domains together are more than half the exam.
AVOID EXAM DUMPS
Sites selling “actual AI-103 questions” offer leaked or invented material. Using them breaks the exam agreement you accept before every Microsoft exam and puts the certification at risk. Many AI-103 dumps are also recycled AI-102 questions that test skills the new exam no longer weighs.
How much does AI-103 cost in total?
The exam is $165 in the United States. Your total depends on how you prepare and whether you pass the first time.
| Route | What you pay | Typical total |
|---|---|---|
| Free route | Exam only; Microsoft Learn and the practice assessment are free | $165 |
| Budget route | Exam + one Udemy course + one practice set | About $200–$250, depending on Udemy’s price that day |
| Portfolio route | Exam + Udacity Microsoft Agentic AI for two months | $165 + 2 × $125 |
| Add a retake | A second attempt is charged at the full exam price | + $165 |
Azure usage is the other cost. Labs that deploy models and search indexes are billed to your subscription, so set a budget alert and delete resources when you finish each exercise. Microsoft’s credential page also lists an Exam Replay offer; check its current terms before you book. Renewing the certification each year is free, and if you fail, Microsoft’s retake policy makes you wait 24 hours before a second attempt and 14 days between later ones, with at most five attempts in 12 months.
AI-103 exam day: what to expect
You can sit AI-103 at a Pearson VUE test center or online with a proctor watching through your camera. Online exams need a quiet room, a clean desk and a webcam; run Pearson VUE’s system test on the same computer and network a day before.
- Budget your time. Plan roughly 1.5 minutes per standard question and keep 20 to 25 minutes for any case study, which may come at the end.
- Use Microsoft Learn sparingly. Look up a detail only when you can name exactly what you need; otherwise flag the question and return to it.
- Breaks have a cost. Microsoft allows breaks, but you cannot return to questions you saw before the break.
- Read for the constraint. Many questions turn on a single requirement such as no API keys, lowest cost or data that must stay in a region.
If you pass, the certification appears in your Microsoft Learn profile, where you can share it and later renew it.
Renewing AI-103
Microsoft associate certifications expire after one year. Microsoft’s renewal page says renewals are free: you pass an unproctored, open-book online assessment on Microsoft Learn before the expiry date. Renewal assessments track changes to the exam, which matters for AI-103 because Foundry changes quickly.
Is the AI-103 certification worth it?
For developers who build AI features on Azure, yes. It is Microsoft’s developer certification for generative AI apps and agents on Azure (AB-100, the agentic business solutions exam, targets Dynamics 365 and Power Platform instead), and the preparation itself is useful work: the labs teach you to deploy, secure and evaluate an agent, the same tasks the study guide lists for the role. At $165 with free preparation available, the cost is low.
It is worth less in three situations:
- You do not work with Azure. The skills transfer, but the certification signals Azure specifically. Compare the AWS machine learning certifications or our Google Cloud certification guide.
- You want a first job in AI with no development experience. A certificate without projects carries little weight; build two or three projects first, or choose a program with reviewed projects.
- You need a broad AI credential rather than a platform one. Our guide to AI certifications compares vendor and university options.
If you build on Azure but not on the AI layer, AZ-104 or AI-200 may be the better first step, and our cloud certifications guide lays out the AWS, Azure and Google paths side by side.
Frequently asked questions
What is the AI-103 certification?
AI-103 is the exam for Microsoft Certified: Azure AI Apps and Agents Developer Associate. It tests whether you can build, deploy and manage generative AI apps and AI agents on Azure with Python and Microsoft Foundry, including retrieval, vision, language and document extraction. It replaced AI-102, the Azure AI Engineer Associate exam, which Microsoft retired in 2026.
How much does the AI-103 exam cost?
$165 in the United States. Microsoft sets the price by the country where the exam is proctored, so it can be higher or lower elsewhere. Renewing the certification each year is free.
Is AI-103 hard?
It is an intermediate exam and most candidates find it moderately hard. It is harder if you have not written Python against Azure services, because questions ask you to reason about SDK code, authentication and configuration. Hands-on practice in Microsoft Foundry is the best way to make it easier.
How long does it take to prepare for AI-103?
Microsoft’s free course for AI-103 is about 30 hours of learning paths. Add hands-on labs and practice tests and most working developers need four to six weeks at 6 to 8 hours a week. Allow more time if you are new to Azure or to Python.
What is the passing score for AI-103?
700 or higher. Microsoft reports a scaled score, so 700 is not the same as answering 70% of questions correctly.
How many questions are on the AI-103 exam?
Microsoft does not publish a fixed number. One candidate who passed in July 2026 reported 63 questions, including a case study of 7 questions at the end, in 100 minutes. Expect roughly 50 to 65 questions.
Is AI-103 open book?
Partly. Microsoft lets you open Microsoft Learn during associate and expert exams, including AI-103. Candidates say the pages load slowly, so use it to confirm a detail rather than to learn a topic during the exam.
Should I take AI-102 or AI-103?
AI-103. Microsoft retired AI-102 and its renewal assessment in 2026, so you can no longer earn or renew the Azure AI Engineer Associate. If you already hold it, it stays valid until its expiry date.
Should I take AI-103 or AI-200?
Take AI-103 if you build generative AI apps and agents. Take AI-200, Azure AI Cloud Developer Associate, if you build the back-end services, data stores and pipelines behind AI applications. AI-200 is the successor to AZ-204.
Does the AI-103 certification expire?
Yes. Microsoft associate certifications expire after one year. You renew by passing a free, unproctored online assessment on Microsoft Learn before the expiry date.
Is there a free AI-103 practice test?
Yes. Microsoft offers a free practice assessment for AI-103 on its AI Skills Navigator site (you need to sign in), and an exam sandbox that shows the question formats. For more questions, paid practice tests on Udemy cost much less than a retake.
Are AI-103 dumps worth using?
No. Dump sites sell leaked or invented questions. Using them breaks the exam agreement every candidate accepts and puts the certification at risk. They are also often copied from AI-102, which tested different skills.
Related guides
- Udacity Microsoft Agentic AI Nanodegree review
- Udacity AI Engineer Using Microsoft Azure review
- AZ-204 retired: what to take instead (AI-200)
- Best AI certifications in 2026
- Best Microsoft Azure courses
- Best cloud certifications: AWS vs Azure vs GCP