Azure AI Engineer Certification: Exam Guide, Cost & How to Prep

You saw "Azure AI Engineer Associate" on a job posting, or a manager said the words "AI-102," and now you are trying to figure out what to actually study before you burn a weekend on it. Here is the thing nobody updated the old guides to tell you. The exam most of those guides teach, AI-102, retired on June 30, 2026. You cannot book it, sit it, or earn the Azure AI Engineer Associate certification through it anymore. The live path is a different, newer exam called AI-103, and it is not a renamed AI-102. It tests different material, on a different platform, at a different level of hands-on depth. Where the old exam leaned on classic machine learning and standalone services, the new one is about whether you can build, deploy, and maintain AI solutions that lean on generative AI and agents.

That single fact reshapes everything about how you prepare, which is exactly why so much of what you will find online is now quietly wrong. This guide gives you five things: what the credential is actually called in 2026 and how the old and new exams differ, whether it is worth your time versus building agent projects, what the current exam tests including the exam domains and the different types of exam questions you will face, a realistic prep timeline calibrated to your Azure starting point, and a study plan built only around resources that match the exam you will really sit. Microsoft certifications in AI have moved faster than almost any other track in the catalog, so knowing which version you are aiming at is the whole game.

Read: Context Engineering: What it Is & Why It's Important for Your AI Usage

What the Azure AI Engineer Certification Is Called in 2026

The Azure AI engineer credential you are looking for has moved. AI-102 and its Azure AI Engineer Associate certification retired on June 30, 2026, and the current exam is AI-103, a distinct credential built around Microsoft Foundry, generative AI, and AI agents.

Here is the mapping, so you can decode whatever wording sent you here.

Read the table this way. If a posting still says "Azure AI Engineer Associate" or "AI-102," the employer means someone who can design, build, and deploy AI solutions on Microsoft Azure. That role has not disappeared. What disappeared is the specific exam that used to validate it. The current credential aimed at roughly the same work is AI-103, and Microsoft has been careful not to call it a like-for-like replacement. It is a separate exam you prepare for on its own terms.

The two are genuinely different. AI-102 spreads across a wide set of standalone Azure AI services. AI-103 assumes you are building on Microsoft Foundry, and it moves generative AI and agents from a small slice of the old exam to its center of gravity. If someone tells you your AI-102 revision transfers straight over, they are wrong on the highest-scoring part of the new exam.

One naming knot worth untangling, because it trips up nearly every study guide. The platform under this credential got renamed twice in twelve months. It launched as Azure AI Studio in November 2023, became Azure AI Foundry at Ignite in November 2024, and became Microsoft Foundry at Ignite in November 2025. The January 2026 product terms made "Microsoft Foundry" the formal name. In that last change, Microsoft also folded the old Azure AI Services brand (itself formerly Azure Cognitive Services) into a suite called Foundry Tools. So when an older resource says "Azure AI Studio," "Azure Cognitive Services," or even "Azure AI Foundry," it is describing an earlier layer of the same platform under a name Microsoft has since retired.

The same renaming happened to individual pieces of the Azure AI portfolio, and this is where legacy guides quietly mislead you. Guides that tell you to leverage Azure Cognitive Services, wire up Azure Cognitive Search, or build a conversational bot with the Microsoft Bot Framework are describing an older generation of Azure services. Azure Cognitive Search is now Azure AI Search, the Cognitive Services APIs now sit inside Foundry Tools, and Bot Framework patterns have given way to agents built in Foundry. The underlying capability often survives under a new name, but if a resource still uses the old labels as if they were current, treat that as a signal it predates the transition.

Here is the decision rule that matters more than any single date. Check the publish or update date on any resource before you trust a word of it. For this credential, date is the single most important quality signal, more than depth, more than reviews, more than production value. Any study guide, Udemy course, or practice bank published before late 2025 is almost certainly teaching AI-102, a retired exam on a platform with a retired name. A guide dated May 2024 documenting a first-try AI-102 pass is a real account of a real exam, just not the one you can sit now.

This is one of the fastest-moving credentials Microsoft offers. Everything below is accurate as of the verification date at the top. Before you book, confirm the current exam code, beta status, and skills outline on Microsoft's credential page, because this is exactly the kind of thing that changes between the day this publishes and the day you read it.

Read: How to Become an AI Specialist

Is the Azure AI Engineer Certification Actually Worth It?

Whether this certification is worth 40 or more hours depends entirely on which of three situations sent you here, and in one of them the honest answer is to skip it. There is also a wrinkle unique to right now: the current exam is in beta, which changes the calculus for some people.

A manager mandated it for a client engagement

Get it. This is the strongest case, and it is not close. Microsoft partners and consultancies need a certified Azure AI engineer on staff to hold partner competencies and to staff client work that contractually requires certified engineers. If your firm's Microsoft partnership status or a specific engagement depends on a certified body, the certificate is the deliverable. A shipped side project does not satisfy a partner requirement. If your employer's paperwork still names "AI-102," flag the retirement to whoever owns the requirement, since the current path is AI-103 and the two are not interchangeable on a compliance checklist.

It is a "preferred qualification" on a posting you want

Conditional. If the employer is a Microsoft partner or a Microsoft-shop enterprise, the certification carries real weight, and you should pursue it. If it is a startup or an AWS-heavy environment listing it as one of five nice-to-haves, a deployed agent project will tell a hiring manager more than a badge will. Look at the employer.

No specific trigger; you are positioning for a future move

The certification probably matters less than a shipped agentic project. A deployed agent with real users, a documented evaluation framework, and an honest writeup of the tradeoffs you made signals applied capability in a way a certificate cannot. Real-world scenarios are what convince a hiring manager you can ship AI applications. Whether your interest is intelligent automation, general AI product work, or deep Azure engineering, a working system beats a credential more often than the reverse. The beta status reinforces this. Until AI-103 reaches general availability, a stable portfolio is the safer bet if you have no external deadline forcing your hand.

Get it if: a client engagement or partner competency requires certified staff, the target employer is a Microsoft partner or Azure-committed enterprise, or you need a forcing function to systematically learn the current Azure AI stack around Foundry and agents.

Skip it (for now) if: you are positioning generally with no external deadline, the employers you are targeting are AWS-heavy or evaluate on portfolio, or you already have a shipped agentic project that demonstrates the same practical skills.

Two factors to build into the decision. First, Microsoft role-based certifications expire and require annual renewal, and renewal happens only inside a six-month window before your certification lapses. Renewal is free and done through a short online assessment rather than a full re-sit, but it is an ongoing obligation. A portfolio project does not expire. Second, because AI-103 is still in beta, Microsoft has not confirmed a general availability date, and the content can shift as early results come in. If a stable, widely supported exam matters more to you than being early, that uncertainty is a legitimate reason to wait for general availability.

What the Exam Actually Tests (Domains, Services, and Logistics)

AI-103 measures five official domains, and the reason older study guides feel off is that they weight the domains the way AI-102 did. Generative AI and agentic solutions now carry the most weight, and Microsoft Foundry is assumed throughout. The knowledge required has shifted with it, and candidates understand the change fastest when they see the current structure mapped to the specific Azure AI services you actually have to learn.

Weightings are drawn from the AI-103 skills outline and can change while the exam is in beta. Confirm against Microsoft's current study guide before you build a plan around these numbers.

You cannot study "implement generative AI and agentic solutions" as an abstraction. You can only study Azure OpenAI Service, building and orchestrating agents in Microsoft Foundry, tool and function calling, and grounding an agent against your own data. That service-level specificity is the whole point. Microsoft's outline gives you the domains, but your study time has to go to the named services under them.

A few shifts from the retired AI-102 are worth naming, because they explain why old material misleads you. Agentic work was a small sliver of AI-102 and is now the heart of the exam. Classical natural language processing staples such as intent recognition and conversational language understanding have largely disappeared, replaced by large language models doing extraction and summarization. Computer vision applications picked up image and video generation on top of the old classification and object detection. Knowledge mining got reframed as grounding, where Azure AI Search feeds retrieval-augmented generation rather than standing on its own. You are still expected to work fluently with the underlying REST APIs and SDKs across all of these, but Python is now the stated development language in the audience profile, where AI-102 was comfortable with either C# or Python.

Do not waste time on retired material. Older practice banks still drill you on services under names from the "formerly" column, or test classic Azure Cognitive Services SDKs at a depth that dwarfs their agent and Foundry coverage. The developer who documented a 2024 AI-102 pass flagged exactly this pattern in one popular practice provider: heavy coverage of retired services, thin coverage of Azure OpenAI. If a practice question is calibrated to an exam that no longer exists, skip it and reallocate the time to Foundry and agents.

Exam logistics for AI-103:

• Format - approximately 40 to 60 questions across multiple choice, drag and drop, and scenario-based case studies, no live coding
• Duration - 120 minutes
• Passing score - 700 out of 1000, a scaled score rather than a straight percentage
• Cost - Microsoft has not published a specific AI-103 beta fee. Standard Azure associate exams commonly list at around 165 USD in the US, and beta exams are sometimes discounted. Price varies by country or region
• Languages - English only during beta
• Practice assessment - not yet released for AI-103 as of mid-2026
• Delivery - online-proctored or at a Pearson VUE testing center
• Documentation access - role-based associate exams give you access to Microsoft Learn inside the exam interface. It is not "open internet," and there is no browser find function, so this is closer to a curated reference than a search engine.

That documentation point is the most useful thing on this list and the one every stale guide buries. Because you can reference Microsoft Learn during the exam, part of what is being tested is how fast you can find the right page. More on how to train for that below.

Real-World Insights From People Who Sat the Exam

Beyond the official outline, the most useful preparation signal comes from people who recently sat the exam. Candidate write-ups and the back-and-forth in an active Azure certification discussion forum describe an exam experience that is very likely to carry into AI-103, since both are role-based associate exams delivered the same way. Confirm the specifics against your own booking, but these lived-experience notes consistently surface details the marketing pages never mention.

The documentation access is real but oversold as a safety net. One candidate who passed with a score of 789 described the in-exam Microsoft Learn search as nearly useless, and warned that if you lean on the docs heavily, you will run out of time. The 120 minutes is enough only if you answer most questions from knowledge and use the docs as a rescue tool for the three to five questions you genuinely cannot resolve. Multiple test-takers independently reported the same thing: time pressure is the real difficulty, and some finished with questions left unanswered.

There is no browser find function inside the exam, and you can keep only a small number of Microsoft Learn tabs open. A coach who trains candidates tells students to cap themselves at roughly 30 seconds on the docs before committing to an answer and moving on. That only works if you already know where things live, which is a skill you build before exam day.

Study beyond the blueprint, and put your focus on hands-on labs rather than passive reading. Experienced candidates noted that questions can assume broader Azure fluency than the skills outline implies, touching networking, containers, and API authentication patterns. Working through the labs in Microsoft Learn builds that fluency far better than notes do. If your Azure administration fundamentals are rusty, refresh them alongside the AI-specific material.

Watch the exam's structure at the very end. One recent test-taker described a final block of a few scenario questions that you cannot return to once you leave it. Slow down deliberately when you reach the last section.

And a logistics note that has nothing to do with knowledge. Read the check-in instructions and prepare your physical space carefully for online proctoring. One candidate had to remove a tissue box from the desk and show the proctor that two loose tissues were clear of writing. Another got logged out mid-exam by a background process on their laptop and lost around ten minutes while the clock kept running. Give yourself buffer time to check in early, and close every background application before you start.

Read: AI Upskilling: Top Firms, Programs, & Tools for Training Your Workforce

How Long It Takes to Prepare (If You Already Know Azure Basics)

The honest answer depends on your starting point, and "already know Azure basics" splits into two very different situations. Find your row and book backward from it. Add extra buffer while AI-103 is in beta, since a practice assessment is not yet available to calibrate your readiness.

The trap for experienced Azure engineers is assuming their existing knowledge transfers to the whole exam. It transfers to most of it. It does not transfer to the newest and least-covered material, which is Microsoft Foundry and the agentic solutions domain. That content did not exist in its current form until recently, so you cannot lean on muscle memory from a project you shipped in 2023.

If you are in row one, spend a disproportionate share of those three to five weeks specifically on Foundry, agents, and grounding. That is where confident engineers lose points, because they scope their prep around the domains they already know and treat the new material as an afterthought. Budget extra time there even when the rest of the exam feels like review.

If you have a hard external date, a manager's deadline, or a client engagement start, work backward using your row. A Python-fluent Azure engineer with an engagement starting in six weeks is comfortably inside the window and should book now, allowing for the fact that beta exams sometimes take longer to return scores. An Azure generalist new to the AI services facing the same six-week deadline is going to be tight and should either start immediately at the top of the hours range or push the date.

The Study Plan: Which Resources Are Current and Which to Skip

Start with one filter and apply it to every resource before you open it. Check the publish or update date first. For this credential, anything predating late 2025 should be treated as potentially misleading. A 2023 or 2024 guide is teaching a retired exam, a different domain weighting, and a platform under two names that no longer apply. That single check will save you more time than any resource on this list will.

Here is the verdict on the resources you are most likely to find.

Microsoft Learn is your spine, and it is where you should start learning before you spend a cent on anything else. It is free, it maps directly to the domains, and its learning paths and Azure AI services documentation are the resources recent passers rate highest by a wide margin. Layer timed practice exams on top of it near the end. Build your sequence around the five domains in order, finishing each before you move on:

1. Plan and manage an Azure AI solution, covering Microsoft Foundry fundamentals, resource setup, and security
2. Generative AI and agentic solutions, covering Azure OpenAI Service and building and orchestrating agents in Foundry, where you should spend the most time
3. Computer vision, covering Azure AI Vision and image generation
4. Text analysis, covering Azure AI Language
5. Information extraction and grounding, covering Azure AI Search and Azure AI Document Intelligence feeding retrieval-augmented generation
6. Readiness gate, the official practice assessment once it exists for AI-103, taken cold as your go or no-go signal. Until it is released, substitute a timed run through a reputable current practice bank and your own built agent as the readiness check.

Now the study move that matters most, and that almost no guide teaches. Because the exam gives you Microsoft Learn access with no browser find function and a running clock, a core tested skill is knowing where things live. You are not memorizing every SDK method. You are building navigation speed under time pressure. So study the docs the way you will use them on exam day. For each domain, learn the layout of that service's documentation. Know where the Azure OpenAI Service quickstarts, the Foundry agent tutorials, the Azure AI Search index configuration pages, and the Document Intelligence model reference actually sit.

The concrete drill: pick a specific task, say calling a specific Azure OpenAI chat completion method with the Python SDK, and time yourself locating the exact method signature in the docs from a cold start, with no find function, the way the exam works. Do this across all five domains until you can reach any service's core SDK reference in under two minutes. Recent passers describe the in-exam search as close to useless and the real winning move as manual navigation you have already rehearsed. That is what makes this drill worth more than another few hundred flashcards. In an exam where the docs are open but the clock is unforgiving and search barely works, rehearsed navigation is the difference between rescuing three uncertain questions and running out of time on them.

The Bottom Line

The hardest part of this certification in 2026 is not the exam. It is aiming at the right target while the target is still being repositioned. AI-102 closed, AI-103 opened in beta, the platform got renamed twice, and most of the study material online is still quietly teaching a version of Azure AI engineering that no longer exists. If you take one thing from this guide, take the habit of checking the date and the exam code on every resource before you trust it, because that single filter separates preparation that pays off from weekends spent studying a retired blueprint.

The deeper signal underneath the naming churn is worth naming plainly. Microsoft moved this credential toward agents, Foundry, and generative AI because that is where the actual work went. The engineers who stand out are not the ones who memorized a service catalog. They are the ones who can build, deploy, and maintain AI solutions that hold up in production, then talk credibly about the tradeoffs they made. Treat the AI-103 skills outline as a list of things to build, sit the exam once you can build them, and let the certificate confirm a capability you already have rather than substitute for one you do not. Pair the badge with a shipped agent and a real evaluation framework, and you become the candidate a hiring manager remembers.

Build the skills behind the certification with Leland

Certifications validate what you can already do, so the fastest way to earn one is to build the real skill first. Leland's AI Builder Program is a live, cohort-based program that turns knowledge workers into people who ship real agents and workflows, taught by operators who do this work. To go deeper on a specific gap, browse Leland's AI Automation & Agents coaches for one-on-one help, or start free with an upcoming live AI agents event and build alongside people doing exactly what AI-103 now tests.

See also: Top 10 AI Consultants and Experts

Read next:

• The 5 Best AI Coding Agents: Pros & Cons, Reviews, & Which is Best for You
• The 5 Best AI Tools & Agents for Business: Reviewed & Ranked (2026)
• The 5 Best AI Tools & Agents for Developers: Reviewed & Ranked (2026)
• The 5 Best AI Voice Agents (By Type & Function) [2026]
• The 5 Best AI Newsletters to Subscribe to in 2026
• The 5 Best AI Tools & Agents for Finance: Reviewed & Ranked (2026)
• The 5 Best AI Tools & Agents for Video Editing: Reviewed & Ranked (2026)
• The 5 Best AI Personal Assistants: Reviewed & Ranked (2026)
• Agentic AI vs. AI Agents: Differences & What You Need to Know

FAQs

Is the Azure AI Engineer certification still called AI-102?

• No. AI-102 and its Azure AI Engineer Associate certification retired on June 30, 2026, and you can no longer book or earn it. The current associate credential aimed at Azure AI engineering is AI-103, which leads to the Azure AI Apps and Agents Developer Associate certification and is built around Microsoft Foundry, generative AI, and agents. If a job posting still says "AI-102," that is the role the employer means, but the live exam is AI-103. Always confirm the current code on Microsoft's credential page before booking.

Are the AI-102 study materials I found online still accurate?

• Mostly no. Any guide, course, or practice bank built for AI-102 targets a retired exam and an older version of the platform, back when it was called Azure AI Studio or Azure AI Foundry, and services sat under Azure Cognitive Services. Those materials under-cover agents, Microsoft Foundry, and Azure OpenAI, and they over-cover services that lost weight or were renamed. Check the publish date first and favor material updated in late 2025 or later that explicitly names AI-103 and Microsoft Foundry.

How long does it take to prepare if I already know Azure?

• If you are already comfortable with Python and working in Azure, budget roughly three to five weeks at eight to ten hours per week, with a disproportionate share aimed at Microsoft Foundry and the agentic content, since that is the newest material and the part your classic Azure knowledge is least likely to cover. Someone starting from near-zero Azure should plan eight to twelve weeks. Add buffer while AI-103 is in beta, because a practice assessment to gauge readiness is not yet available.

Is the Azure AI Engineer certification worth it, or should I just build agent projects?

• If a manager needs certified staff for a client engagement or a Microsoft partner competency, get it, because it is effectively required and a portfolio project will not satisfy that requirement. If it is a preferred qualification or general positioning, a deployed agentic project with a documented evaluation framework often signals more than the badge alone. The certification carries the most weight inside Microsoft-partner consultancies and Azure-committed enterprises.

What does the Azure AI Engineer exam actually test?

• AI-103 measures five domains: planning and managing an Azure AI solution, implementing generative AI and agentic solutions, computer vision, text analysis, and information extraction. In practice, that means learning Microsoft Foundry, Azure OpenAI Service, Azure AI Vision, Azure AI Language, Azure AI Search, and Azure AI Document Intelligence, with agents and generative AI weighted most heavily and Python as the stated language.

Do I need AI-900, AI-901, or AZ-900 before the Azure AI Engineer certification?

• No, there is no formal prerequisite. Azure AI Fundamentals is a separate, beginner-level certification and not a required stepping stone. Note that AI-900 also retired on June 30, 2026, and its fundamentals successor is AI-901. If you already know Azure basics and Python, you can prepare for the associate-level exam directly without a fundamentals cert first, though the fundamentals path is a reasonable first rung if the AI services are new to you.

What is the passing score, and how much does it cost?

• The passing score is 700 out of 1000, a scaled score rather than a straight percentage. Microsoft has not published a specific fee for the AI-103 beta. Standard Azure associate exams commonly list around 165 USD in the US, and beta exams are sometimes offered at a reduced rate, with price varying by country. Confirm both against Microsoft's current exam page before booking, since beta details change.

Does the Azure AI Engineer certification expire?

• Yes. Microsoft role-based certifications expire annually and require renewal, but renewal is free and completed through a short online assessment rather than retaking the full exam. You can renew only during the six-month window before your certification lapses. Factor this ongoing maintenance into your decision if you are weighing the certification against a one-time portfolio project.