AI-200 vs AI-103: Which Azure AI Developer Certification Fits You Now That AI-102 Is Retired

Microsoft's AI-102 page now carries a retirement warning. Two associate credentials cover the ground it held. AI-200 tests the Azure back end an AI application runs on: containers, data services, messaging, monitoring. AI-103 tests the AI itself, in Python and Microsoft Foundry. Choose by which half of that split you already build.

What happened to AI-102

The certification AI-102 led to was Azure AI Engineer Associate. Its credential page now opens with a warning box, which states that the certification is retired, and the renewal assessment with it. A link from that box points to a post on Microsoft's tech community blog.

The warning gives no date. The page does carry a last-updated stamp of 23 December 2025. That is metadata about the page, not the day the credential retired, and we will not present it as one.

Two other lines say the same thing in a quieter way. Where the preparation section would normally list courses, the page reads: "No training available for this exam." The exam behind it ran for 100 minutes and was published in ten languages. That is now a description of something you cannot book.

Nowhere on any page we read does Microsoft name a successor. So we will not name one either. What we can do is show you the evidence and let you weigh it: AI-102's six skill areas and AI-103's five line up almost word for word.

The six areas AI-102 was assessed on	Where that wording turns up on AI-103
Plan and manage an Azure AI solution	Unchanged. The same seven words open AI-103's list
Implement generative AI solutions	Merged. AI-103 runs generative and agentic work as a single area
Implement an agentic solution	Merged into that same area, which reads implement generative AI and agentic solutions on the newer page
Implement computer vision solutions	Unchanged
Implement natural language processing solutions	Renamed. The newer list calls it implement text analysis solutions instead
Implement knowledge mining and information extraction solutions	Shortened. Knowledge mining drops out, and implement information extraction solutions is what remains

Now compare that with AI-200, whose four areas are containers, Azure data management services, messaging and serverless, and security and observability. Not one of them appears on either list above. That is the whole point of this article, and it saves you reading two syllabuses:

AI-200 is the back end the AI runs on. AI-103 is the AI.

They are not two tiers of one credential, and they are not really competitors. They divide the territory AI-102 used to hold in one piece.

AI-200: the Azure back end an AI application runs on

Microsoft files AI-200 under the Developer role, at Intermediate level, with Azure as its product. Two subjects are listed against it: Application development and Artificial intelligence. The certification it leads to is Azure AI Cloud Developer Associate.

Read the metadata again, because it is doing real work. Application development is on this one and on neither of the others. That is the seam.

Underneath AI-200's four skill areas are services a platform engineer already knows by name. On the storage side: Cosmos DB for NoSQL, Azure Database for PostgreSQL, Azure Managed Redis. On the hosting side: Azure Kubernetes Service and Azure Container Apps. Then the services that connect and protect them: Service Bus, Event Grid, Functions, Key Vault, and KQL queries over your logs. The full outline, with the weight attached to each area, is in our AI-200 exam guide.

The audience profile assumes you arrive with Python programming, with containerized applications on Azure, and with vector databases already understood. Those are entry conditions, not lessons. We take the whole profile apart, item by item, in our AI-200 study plan.

On format, the page allows 120 minutes, warns that the sitting is proctored, and says interactive components may appear. Booking is open through Pearson VUE for exam AI-200. There is no worldwide fee: the amount tracks the country or region of your test centre.

And one line changes who can sit it at all. AI-200 is offered in English, and the page lists no second language.

AI-103: the AI itself — Python, Microsoft Foundry and agents

AI-103 leads to the Azure AI Apps and Agents Developer Associate credential. Its metadata differs from AI-200's in three places at once. Two roles are listed, AI Engineer and Developer, where AI-200 lists one. One subject is listed, Artificial intelligence, where AI-200 lists two. And a second product joins Azure: Microsoft Foundry. The level is Intermediate, the same as AI-200.

The certification's own one-line description puts advanced Azure AI solutions at the centre — designed, built and deployed with Python and Microsoft Foundry.

Five skill areas are assessed. Planning and managing an Azure AI solution comes first. Then generative AI and agentic work, together in one area. Then three application areas: computer vision, text analysis, and information extraction.

The overview page expects you to build applications in Python already, and to know your way around general AI, generative AI, and Azure's services. The job it describes is a collaborative one. You work with data scientists and solution architects, with DevOps engineers and cloud security engineers, and with the business stakeholders who asked for the thing.

Format matches AI-200 closely. You get 120 minutes, proctored, with interactive components possible, booked through Pearson VUE for exam AI-103, at a price set by country or region.

Then the difference that matters most for a global reader. AI-103 is offered in ten languages. In alphabetical order: Chinese (Simplified), Chinese (Traditional), English, French, German, Italian, Japanese, Korean, Portuguese (Brazil) and Spanish. Microsoft's page prints them in a different order; the set is the same.

Official practice material: one has it, one does not

AI-103's page links a practice assessment hosted on AI Skills Navigator. You need to be signed in to launch it, and Microsoft describes the experience there as newly upgraded.

AI-200's page has no such link. It states that no practice assessment is currently available for this exam, and it explains the absence with a schedule rather than a date. Practice assessments normally follow about eight weeks behind an exam's move out of beta into general availability. The page does not say where AI-200 stands on that timeline. Neither will we, because no page we read settles it.

Both pages do link the free exam sandbox, which shows you the software and the question formats. It contains no subject matter for either exam.

Our commercial position, stated here rather than left for you to notice. We sell AI-200 practice exams. That gives us an interest in one side of this comparison. This is the section where such an interest would show, so you should know about it while you read the paragraph above. We publish no AI-103 material at all. For AI-103, Microsoft's own practice assessment is free and is the first thing we would use.

The decision, by what you already do

Skip the syllabuses and describe your last month of work instead.

AI-200 fits you if the week looks like this. You write Dockerfiles and ship images. You keep things running on Azure Kubernetes Service or Azure Container Apps. You tune a query and watch what it costs. You put an embedding into a database and get similar rows back out. You wire a queue to a function, keep secrets in Key Vault, and go to the logs when a request fails. The models are somebody else's problem. The platform that serves them, quickly and visibly, is yours.

AI-103 fits you if the week looks like this. You write Python against AI services. You build and manage agents in Microsoft Foundry. You put a vision, text or extraction step inside an application and worry about what comes back. The container it runs in is somebody else's problem; what the model does is yours.

Two more considerations decide the remaining cases, and both are honest constraints rather than advice.

Language. If you cannot sit an exam in English, AI-200 is not currently an option and AI-103 is. The AI-200 study guide does record a Microsoft policy that softens this a little. Where a preferred language is missing from an exam, candidates may request 30 extra minutes. Ask for it when you book.

Job titles in your market. AI-103 carries the AI Engineer role that AI-102 carried, and that is the phrase most job adverts still use. AI-200 carries Developer. If a specific advert is the reason you are certifying, read the advert's words and match them. That is our suggestion, not a Microsoft recommendation.

If both descriptions fit, take both. They barely overlap, so the second one is mostly new work rather than a rerun of the first.

Our AI-200 practice exams follow the published outline for that exam, and every question links the official Microsoft page behind it. If you are still deciding what AI-200 covers, start with the exam guide. If you want to know what is and is not available for it yet, we track that separately.

Independent study material. ciply.io is not affiliated with, endorsed by, or sponsored by Microsoft.